Vr interaction method and system for virtual disassembly and assembly teaching of three-electric system of new energy vehicle

Abstract

The application discloses a VR interaction method and system for virtual disassembly and assembly teaching of a three-electricity system of a new energy vehicle; the application relates to the technical field of VR teaching; a 3D model and a scene configuration file are loaded, a virtual scene is updated in real time, a scene topology structure is constructed and component relationships are recorded, and safety rules and process configurations are initialized. The 3D model and the scene configuration file are loaded using GPU-based asynchronous loading technology; the virtual scene is updated in real time and information is prompted; the parent-child hierarchical relationships of components are recorded through a tree data structure, and the disassembly and assembly process is controlled based on a constructed behavior tree, so that students can operate in the correct order and steps. Safety rules and process configurations, such as the fact that high-voltage electricity must be disconnected before a battery is disassembled, are built in, and safety accidents caused by misoperation are effectively avoided. When a student's operation is irregular, the application triggers a state machine to roll back to the most recent legal state, helps the student correct errors in a timely manner, and ensures the smooth progress of the disassembly and assembly process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of VR teaching, specifically to the direction of automobile disassembly teaching and VR technology, and particularly to a VR interaction method and system for virtual disassembly teaching of a new energy vehicle three-electric system. BACKGROUND

[0002] Virtual reality technology (VR) is an interactive technology that generates a three-dimensional dynamic simulation environment through a computer to provide an immersive experience for users. VR technology can simulate a real new energy vehicle workshop scene, including vehicles, tool carts, part carts, and protective equipment. Students can freely move around in the virtual scene, observe the overall vehicle structure and various components, and interact with the scene through a handle to bring an immersive disassembly experience. At the same time, students can choose any task on a tablet for practical exercise and self-checking, and operate according to standard operation steps.

[0003] However, the prior art has some defects:

[0004] CN108711326A discloses an automobile education and training system based on virtual simulation (publication date: October 26, 2018); CN119477615A discloses a virtual training method for automobile vocational education (publication date: 2025-02-18); CN118197120A discloses a new energy vehicle disassembly process teaching demonstration system based on VR interaction technology (publication date: 2024-06-14); such prior art simply builds a VR demonstration scene and demonstrates automobile disassembly teaching according to a fixed demonstration process, and students cannot participate in it and can only passively understand the visualized knowledge, with weak actual interaction effect and unsatisfactory teaching effect.

[0005] CN113744590A discloses a VR interactive teaching device based on virtual disassembly and detection of the high-voltage part of a pure electric vehicle (publication date: 2021-12-03); CN108806385A discloses a VR immersive multi-person collaborative automobile teaching training system and teaching method (publication date: 2018-11-13); such prior art can realize the combination of VR demonstration and disassembly interaction, allowing students to interact with the VR scene and simulate the disassembly of automobile components, but the structure of the new energy vehicle three-electric system is complex, the body is large, and some specific mechanisms must strictly follow a specific disassembly process (such as “battery pack → module → electric core”); the above traditional technology does not add necessary constraints, so students cannot learn the strict disassembly process.

[0006] The defect of weak constraint effect further leads to another technical problem: the screw disassembly of the three-electricity system of new energy vehicles must strictly follow a specific torque, because in the battery system, improper torque can cause the battery connection to be loose or damaged, affecting the performance and life of the battery. For example, when disassembling the battery connection bolt, excessive torque can damage the battery connector, causing poor contact or short circuit. The bolt connection in the motor and the electric control system also requires precise torque control. Improper torque can cause internal parts of the motor to shift, exacerbate wear and tear, or cause poor connection of the electric control system, affecting the performance and reliability of the entire three-electricity system. Traditional technology cannot achieve the corresponding conditional constraints, cannot simulate the selection of torque and the tactile sensation brought by actual application, cannot cultivate students' muscle reflexes for the appropriate force size under various torque conditions, and cannot effectively cultivate students' practical experience.

[0007] To this end, the present application proposes a VR interaction method and system for virtual disassembly and assembly teaching of the three-electricity system of new energy vehicles. SUMMARY

[0008] Therefore, the present application hopes to provide a VR interaction method and system for virtual disassembly and assembly teaching of the three-electricity system of new energy vehicles to solve or alleviate the technical problems existing in the prior art, i.e., how to further provide necessary constraint functions for each process demonstration item, achieve the purpose of process teaching, further introduce a disassembly torque simulation mechanism to simulate the torque constraint required for disassembling the three-electricity system in practice, and thus cultivate students' hand operation ability and muscle reflex, and at least provide one beneficial choice; the technical solution of the present application is as follows:

[0009] In a first aspect, the VR interaction method for virtual disassembly and assembly teaching of the three-electricity system of new energy vehicles is as follows:

[0010] The present application aims to build a new energy vehicle education and training system through virtual reality (VR) technology, realize immersive disassembly experience and teaching evaluation feedback. The scheme includes four core parts: resource loading and scene construction, user interaction processing, disassembly process logic control, and teaching evaluation and feedback. By loading 3D models and scene configuration files, a real virtual workshop environment is constructed; the user's operation intention is captured to realize interaction with virtual components and simulate real disassembly feel. Based on the definition of disassembly steps in a directed acyclic graph, the process logic is ensured to be accurate, and illegal operations are monitored in real time. At the same time, student operation data is recorded to provide intelligent evaluation feedback and fault simulation troubleshooting training. This scheme effectively solves the problems of weak interactivity and insufficient disassembly process constraints in traditional automobile education and training, and improves the teaching effect and students' learning experience.

[0011] (II) Technical solution

[0012] 2.1 Step S1, system initialization and scene loading:

[0013] Load 3D models and scene configuration files, update virtual scenes in real time, build scene topology and record component relationships, initialize safety rules and process configuration.

[0014] 2.1.1 Step S100, resource loading and model instantiation:

[0015] Load 3D models and scene configuration files using GPU-based asynchronous loading technology; update virtual scenes in real time and prompt information, achieve component outline light through Shader programming, and generate floating UI in VR screen space; support VR split screen display using multiple camera viewports.

[0016] 2.1.2 Step S101, start model LOD level management:

[0017] Render the model only in front of the relative camera viewport (close distance) to reduce GPU load;

[0018] 2.1.3 Step S102, scene topology construction:

[0019] Record the parent-child hierarchical relationship of components through tree data structure (for example, the parent-child hierarchical relationship of the power supply system of a car is "battery pack → module → battery cell"), bind quality attributes and collision volume attributes, and use quaternions and matrix transformation to manage component pose (Position & Rotation).

[0020] 2.1.4 Step S103, initialize safety rules and process configuration:

[0021] Based on the constructed behavior tree control process, parse the disassembly sequence and safety constraints from the configuration file (for example, when disassembling the battery, the constraint condition is "must first disconnect the high-voltage power").

[0022] 2.2 Step S2, user interaction processing:

[0023] Capture and analyze user's operation intention in VR environment, generate haptic feedback to simulate real disassembly feeling, simulate collision response through rigid body dynamics to ensure the correctness and safety of disassembly operation, and trigger state machine rollback to correct illegal operation.

[0024] 2.2.1 Step S200, handle / gesture input analysis:

[0025] The handle detects the component pointed by the user through ray casting and calculates the interaction point coordinates; the gesture uses the MediaPipe library to identify hand key points, including grabbing gestures and rotating gestures.

[0026] The method of ray casting includes:

[0027] S2000, Ray r(t) generation: The handle ray is defined by a position O (origin) and a direction d (unit vector): t' ≥ 0; where t' is the parameter of the intersection of the ray with each plane;

[0028] S2001, Collision detection (AABB bounding box intersection test): For each component's Axis Aligned Bounding Box (AABB) in the scene, set: ;

[0029] S2002, Calculate the entering parameter t enter and the exiting parameter t exit of the intersection of the ray with each plane:

[0030] ;

[0031] where x, y, z represent the coordinate axes, and min represents the minimum parameter thereof;

[0032] S2003, Calculate the intersection condition: ;

[0033] S2004, Take the nearest legal intersection parameter t hit , and calculate the intersection point coordinates:

[0034] ;

[0035] The intersection point coordinates are: ;

[0036] 2.2.2 Step S201, tool-component interaction:

[0037] Detect the collision bounding box (AABB) of the handle and the component, trigger the grab event; determine whether the current tool is applicable through the Drools-Rete algorithm (such as “whether the torque wrench matches the bolt specification”).

[0038] S2010, Collision bounding box detection: Based on the handle grab point coordinates P hit = (x, y, z) (from the ray projection result), the minimum / maximum corner points b min , b max of the component AABB bounding box are:

[0039] ;

[0040] S2011, Determine the collision condition as: ;

[0041] S2012, Determine the trigger grab event GrabEvent:

[0042] ;

[0043] S2013, tool applicability judgment (Drools-Rete algorithm):

[0044] Current tool attributes TT = (type, min torque , max torque ); type is the code of the tool model, min torque and max torque are the minimum and maximum torque parameter encodings of the tool respectively;

[0045] Target bolt size BB = (diameter, required torque ); diameter is the diameter parameter code of the bolt, required torque It is the torque parameter code required for the bolt;

[0046] Rule matching (Rete network condition evaluation) is a logical expression:

[0047] ;

[0048] In S2014, the Rete algorithm was executed: matching tool types by nodes and associating bolt requirements, and finally activating the rules:

[0049] ;

[0050] Where n is the number of nodes, i is the node index; applicable is the rule that is activated;

[0051] 2.2.3 Step S202: generating tactile feedback:

[0052] The torque required to tighten the screw is calculated based on the physics engine, and the vibration motor in the handle is driven by a waveform generation algorithm (Sine / PWM).

[0053] The implementation of this step involves a tactile feedback generation algorithm, the process is as follows:

[0054] S2020, mapping of torque to vibration intensity: Assume that the current torque calculated by the physics engine is T (unit: Nm), and define the proportional relationship between the vibration amplitude A and the torque T ;

[0055] Where k is the gain coefficient, T max The maximum torque allowed by the system needs to be calibrated according to the performance of the handle motor;

[0056] S2021, generating a sine wave vibration signal V related to the torque sine (t):

[0057] ;

[0058] where f is the reference frequency, set to the haptic sensitive frequency band (e.g. 50-200 Hz); φ is the initial phase; a non-linear mapping term A∝T 2 Enhanced high-torque feedback.

[0059] The interaction point coordinate P hit Adjust the vibration space attenuation (e.g. the farther the weaker vibration).

[0060] S2022, PWM (Pulse Width Modulation) vibration signal generation:

[0061] Define the relationship between the duty cycle D∈[0,1] and torque (linear mapping) ;

[0062] S2023, PWM waveform V PWM (t) in the period T PWM is expressed as:

[0063] ;

[0064] S2024, vibration signal superposition and output, the final driving signal V(t) mixed two kinds of waveform:

[0065] ;

[0066] where α and β are weight coefficients;

[0067] 2.2.4 Step S203, rigid body dynamics simulation:

[0068] Use the physical engine to solve the Newton-Euler equation, calculate the component motion, and use space partition to screen the object pairs that may collide; use the Separating Axis Theorem (SAT) algorithm to detect collisions, and use the impulse method to calculate the post-collision speed.

[0069] S2030, space partition screening collision pairs:

[0070] S20300, BVH tree construction, recursively executed on all objects in the scene:

[0071] Calculate the bounding box;

[0072] Divide the object set into left and right sub-trees along the longest axis;

[0073] The termination condition is that the leaf node contains ≤k' objects;

[0074] S20301, traverse the BVH tree to screen potential collision pairs: For any two nodes N1 and N2, if their bounding boxes intersect, recursively check their child nodes and screen out CPL. The screening method is as follows:

[0075] ;

[0076] S2031, Separating Axis Theorem Collision Detection:

[0077] Projection calculation: for all candidate separation axes a of two objects A and B i (including surface normal and cross product edge direction), calculate the projection interval Proj A and Proj B :

[0078] ;

[0079] ;

[0080] Where q and p are the face normal and the cross product edge, respectively.

[0081] The separation condition is: if there is any axis a i satisfy or , then the objects do not collide, otherwise a collision occurs.

[0082] S2032, Impulse Method Collision Response:

[0083] Normal impulse calculation: Assume collision normal vector n, restitution coefficient e, relative velocity v rel = v A -v B , impulse ;

[0084] Among them, r A 、r B is the vector from the collision point to the center of mass; v A and v B They are the observation speed and the observed speed respectively. Then update the speed after collision: ;

[0085] 2.2.5 Step S204: Triggering state machine rollback:

[0086] Graph theory is used to detect disassembly and assembly sequence violations (such as directly separating components without removing bolts), triggering state machine rollback.

[0087] S2040, Graph Theory Modeling of Assembly and Disassembly Process: Let the assembly and disassembly process be a directed graph G=(V,E), where the vertex set V={v1,v2,...,v n} represents a legal operation state (such as "power off", "bolt has been removed"); edge set E V i ×V j represents the allowed state transition (such as );

[0088] S2041, current state path tracking: define state history stack S = [s0, s1,..., s t'' ], where s K ∈ V represents the state of the Kth step, and the current state is s t'' ;

[0089] S2042, when the user performs an action a, the state transition is triggered, if the transition is not legal: (s t ,s'')∉E, then trigger the rollback condition RollbackTrigger, and then trigger the rollback condition RollbackTrigger;

[0090] S2043, rollback target state determination: find the nearest legal predecessor state s p by reverse searching the history stack S, so that ∃(s p ,s valid )∈E, where s valid is the legal state of the current scene.

[0091] S2044, after rollback, the state is updated to S←[s0, s1,..., s p ].

[0092] 2.3 Step S3, disassembly flow logic control:

[0093] Based on the directed acyclic graph (DAG), define the disassembly steps, use breadth-first search (BFS) to push the current executable operation (such as "the next step should disconnect the high-voltage interface"); at the same time, through the rule engine, monitor the illegal operation in real time, and provide an alarm;

[0094] Use a hash table (Hash Table) to map the association between tools and actions (such as "electric wrench → tighten bolt"), and match the user's intention through cosine similarity.

[0095] 2.3.1 Step S300, DAG-based disassembly step pushing:

[0096] Based on the directed acyclic graph (DAG), define the algorithm of this step.

[0097] Let the disassembly flowchart G = (V, E), where the vertex set V = {v1, v2,..., v m} represents a total of m operation steps v i(such as v i = "Power off", v j = "Remove the shell"); Edge set E V×V represents the step dependency (such as E(v i ,v j )∈E means that you must first "turn off the power" before "removing the shell");

[0098] 2.3.2 Step S301, generating the current executable step set (BFS traversal):

[0099] Define the adjacency matrix A, where A ij = 1 if and only if E(v i ,v j )∈E. From the current set of completed steps S done Start by finding all directly reachable steps through BFS:

[0100] ;

[0101] Executable actions prompted by the user interface ;

[0102] The priority order is preset based on the security level or topological order.

[0103] 2.3.3 Step S302: Rule Engine Violation Monitoring:

[0104] Set the current operation to a t" , corresponding to state transition , check the validity function AlarmTrigger:

[0105] ;

[0106] Then, when an alarm is triggered, it is forced to roll back to the most recent legal state (see the state machine rollback algorithm above).

[0107] 2.3.4 Step S303: Tool-action mapping and intent matching:

[0108] The hash table stores the tool-action relationship and defines a hash function H:KK→V, where the key KK is the tool type (such as "electric wrench") and V is the set of allowed actions, such as {tighten the bolt, remove the nut};

[0109] Matching user intentions based on cosine similarity: Let the feature vector of the user's current operation u=(u1,u2,...,u m ) (including sensor data such as torque and angle), the target action requires vector v a =V=(v1,v2,...,v m ), then the matching degree ; where ||·|| denotes the logical operator "or".

[0110] Selecting optimal actions ;

[0111] 2.4 Step S4, Teaching Evaluation and Feedback:

[0112] Timestamp, operation type, and component ID information are recorded and stored as a time-series database (InfluxDB).

[0113] Based on the weighted method as the scoring rule, randomly inject faults (such as "communication interruption of the electric control system"), and use decision trees (ID3 / C4.5) to determine whether the student's troubleshooting path is correct.

[0114] 2.4.1 Step S400, Operation Data Collection and Time-Series Storage:

[0115] When the user performs key operations, including grabbing components, using tools, and completing steps, data recording is triggered; recorded fields include:

[0116] (1) Timestamp: accurate to milliseconds (e.g., 2024-05-20T14:23:45.678Z);

[0117] (2) Operation Type: categorical label (e.g., disassembling battery shell, incorrect non-insulated operation);

[0118] (3) Component ID: unique identifier (e.g., a certain block of battery marked as Battery.Module3.Cell25);

[0119] (4) Additional Data: operation parameters, tool model, and time consumption.

[0120] Time-series database (InfluxDB) stores information including:

[0121] (1) Data Sharding: partitioned by student ID or operation stage, accelerating queries.

[0122] (2) Data Compression: using Delta Encoding for repeated operations (mainly for consecutive screwing).

[0123] (3) Tag Indexing: inverted index for high-frequency query fields (such as component ID, error type).

[0124] 2.4.2 Step S401, Data Stream Pipeline Execution:

[0125] Real-time stream processing is performed through Kafka or MQTT to transmit data from the VR client to the database;

[0126] De-duplication and verification: filter invalid operations (e.g. false triggers caused by handle shaking).

[0127] 2.4.3 Step S402, diagnostic logic of fault simulation:

[0128] Based on predefined common faults (such as interruption of electric control communication, short circuit of battery monomer), randomly inject faults after user performs specific operations (such as disassembling electric control shell); trigger faults with probability p every fixed time interval (such as 5 minutes); and simulate virtual component abnormalities (such as instrument panel display error code, component smoke special effect).

[0129] 2.4.4 Step S403, evaluation of student troubleshooting path (decision tree logic):

[0130] The order of student troubleshooting steps (such as checking power supply first, then testing signal line);

[0131] Whether to use the correct detection equipment (such as multimeter to measure voltage);

[0132] Response time is the delay from the occurrence of the fault to the start of troubleshooting;

[0133] 2.4.4 Step S403, decision tree training (ID3 / C4.5):

[0134] Select the feature with the highest information gain (such as "whether to prioritize detection of master module") from the "correct / incorrect" troubleshooting path marked in historical data; compare student operation with optimal path generated by decision tree in real time.

[0135] If the student continuously deviates from the correct path, trigger a prompt (such as "suggest checking CAN bus connection"); based on path matching degree and troubleshooting efficiency, calculate the score based on weighted method.

[0136] (Three) Mechanism to solve technical problems:

[0137] (1) Enhance interaction effect: detect the component pointed by the user through ray projection, calculate the interaction point coordinates, realize accurate handle and gesture operation. Students can freely walk in the virtual scene, observe the whole vehicle structure and components, and perform actual interactive operation. Detect the collision bounding box of the handle and the component, trigger the grabbing event, and determine whether the current tool is suitable through Drools-Rete algorithm. This mechanism increases the realism and interactivity of the operation.

[0138] (2) Introduce disassembly process constraints: Record the parent-child hierarchical relationship of components through a tree data structure, and bind quality attributes and collision volume attributes. This provides students with a clear disassembly path and hierarchical relationship. Based on the construction behavior tree control flow, parse the disassembly sequence and safety constraints from the configuration file. For example, when disassembling the battery, the system will constrain the condition that "high-voltage power must be disconnected first", ensuring that students operate in the correct order. Through graph theory, detect disassembly sequence violations and trigger the state machine to roll back to the nearest legal state. This avoids disassembly failure or equipment damage caused by student misoperation.

[0139] (3) Simulate torque control: According to the torque calculated by the physics engine, drive the handle vibration motor through waveform generation algorithms (such as Sine / PWM). This mechanism simulates the selection and touch of torque in the actual disassembly process, cultivating students' hand operation ability and muscle reflex. Use the physics engine to solve Newton-Euler equations to calculate the collision response after component movement. This further enhances the realism and credibility of the virtual scene.

[0140] Secondly, the VR interaction system for virtual disassembly teaching of new energy vehicle three-electric system, as shown in Figure 9

[0141] (1) Three-dimensional model and scene management module:

[0142] 3D model library: Stores three-dimensional models of battery packs, motors, electric control systems, and their sub-components (battery cells, BMS, inverters, and cooling systems).

[0143] Scene dynamic loader: Supports switching between different vehicle models (pure electric / hybrid) and different disassembly scenes (complete system disassembly, single component replacement).

[0144] Hierarchical structure display: Supports transparent / semi-transparent display of component layer-by-layer decomposition (battery pack → module → battery cell).

[0145] (2) Physical simulation and interaction module:

[0146] Rigid body dynamics engine: Simulates the mechanical interaction between tools (wrench, lifting equipment) and components (screwing, disassembling gravity influence).

[0147] Collision detection module: Ensures that the disassembly sequence conforms to reality (e.g., do not disconnect high-voltage power before disassembling the battery, and do not separate components before fully removing the bolt).

[0148] Haptic feedback system: Simulates tool operation vibration through VR handle (e.g., torque feedback when tightening screws).

[0149] (3) Disassembly process logic control module:

[0150] ​Step-by-step procedure engine: Step-by-step prompts operations (e.g. power off -> remove the shell -> disconnect the high-voltage interface), and detects illegal operations (e.g. touching high-voltage components without wearing insulated gloves) and triggers warnings / rollbacks.

[0151] Tool matching system: Based on the current step performed by the step-by-step procedure engine, automatically recommend suitable tools (e.g. torque wrench specifications) or prompt incorrect tool usage.

[0152] (4) Teaching and evaluation module:

[0153] Knowledge labeling module: Displays the name, function and parameters (e.g. voltage, torque value) when hovering / clicking on components.

[0154] Fault simulation module: Implant common fault scenarios (e.g. battery leakage, motor overheating) for students to troubleshoot.

[0155] Operation scoring module: Records operation time, error times and tool usage rationality; generates disassembly reports (e.g. missed steps, safety hazards).

[0156] (5) VR device, processor and memory connected to the processor, the memory storing program instructions;

[0157] The processor is connected with the above-mentioned three-dimensional model and scene management module, physical simulation and interaction module, disassembly process logic control module, teaching and evaluation module and VR device;

[0158] The user interface is provided by the VR device;

[0159] The program instructions are executed by the processor, so that the processor executes the VR interaction method as described above.

[0160] Compared with the prior art, the beneficial effects of the present application are:

[0161] 1. Strengthen disassembly process learning and practice: The present application records the parent-child hierarchical relationship of components through a tree data structure, and controls the disassembly process based on the constructed behavior tree, ensuring that students operate in the correct order and steps. Built-in safety rules and process configurations, such as disconnecting high-voltage power before disassembling the battery, effectively prevent safety accidents caused by misoperation. When the student's operation is illegal, the present application triggers the state machine to roll back to the most recent legal state, helping students correct errors in time and ensuring the smooth progress of the disassembly process;

[0162] 2. Simulate real torque control: According to the torque calculated by the physical engine, the handle vibration motor is driven by the waveform generation algorithm to simulate the tactile feedback under different torques in the actual disassembly process, helping students better understand and master the torque control skills. By simulating the real disassembly feeling, the application can cultivate the students' hand operation ability and muscle reflex, and lay a solid foundation for future actual operation;

[0163] 3. Improve teaching evaluation and feedback efficiency: The application can record the operation data of students in real time, including timestamp, operation type and component ID and other information, which provides detailed data support for subsequent teaching evaluation. The application also supports random fault injection to simulate real maintenance scenarios, requiring students to troubleshoot and repair according to the knowledge learned, further exercising the students' practical ability and problem-solving ability. BRIEF DESCRIPTION OF DRAWINGS

[0164] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0165] Figure 1 The method flowchart of the present application;

[0166] Figure 2 The sub-step schematic diagram of steps S200-S203 of the present application;

[0167] Figure 3 The VR effect visualization schematic diagram of the present application;

[0168] Figure 4 The VR disassembly and torque simulation schematic diagram of the present application;

[0169] Figure 5 The physical collision effect schematic diagram of the present application for disassembling the three-electric system shell;

[0170] Figure 6 The plate group effect visualization schematic diagram of the present application after disassembling the three-electric system shell;

[0171] Figure 7 The VR state machine rollback schematic diagram of the present application (with red flashing special effect as prompt);

[0172] Figure 8 The vibration simulation effect comparison schematic diagram of the present application;

[0173] Figure 9 The system composition schematic diagram of the present application. DETAILED DESCRIPTION

[0174] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. The following description sets forth many specific details to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art can make similar improvements without violating the scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0175] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. The devices disclosed in the embodiments are described briefly because they correspond to the methods disclosed in the embodiments. For relevant details, refer to the method description.

[0176] Example 1: Traditional technology often simply builds a VR demonstration scene, and students can only passively understand the visualized knowledge according to a fixed demonstration process, lacking actual operational interaction. The three-electric system of new energy vehicles has a complex structure, and specific disassembly processes must be strictly followed during disassembly and assembly. However, traditional technology does not add necessary constraints to these process steps, resulting in students being unable to learn strict disassembly and assembly processes. During the disassembly and assembly of new energy vehicles, screw removal must strictly follow a specific torque. Inappropriate torque may cause the battery connection to loosen or be damaged, affecting battery performance and life. Traditional technology can neither achieve the corresponding conditional constraints nor simulate the tactile feel brought about by the selection of torque and actual application.

[0177] To do this, see Figure 1 ,This embodiment discloses a VR interactive method for virtual disassembly and assembly teaching of the three-electric system of new energy vehicles;

[0178] In this embodiment, regarding step S1, system initialization and scene loading: loading 3D models and scene configuration files, updating the virtual scene in real time, building the scene topology and recording component relationships, and initializing safety rules and process configuration.

[0179] Specifically, step S100, resource loading and model instantiation: use GPU-based asynchronous loading technology to load 3D models and scene configuration files; update the virtual scene in real time and prompt information, generate a floating UI in the VR screen space; use multi-camera viewport (Multi-Viewport) or single-pass stereo rendering (Single-Pass Stereo) to support VR split-screen display, ensure that the left and right eye views are correctly separated, and achieve an immersive experience.

[0180] Specifically, step S101, start the model LOD (Level of Detail) management: make the model only render in the relative camera viewport close distance, reduce the GPU load; when the model is close to the camera, use high detail level rendering; when the model is far away from the camera, switch to low detail level to reduce the GPU load. By presetting different LOD models and dynamically switching according to the distance in runtime, efficient use of resources is realized.

[0181] Specifically, step S102, scene topology structure construction: record the parent-child hierarchical relationship of components through tree structure (for example, the parent-child hierarchical relationship of the power supply system of the car is "battery pack → module → battery"), and bind the quality attribute and the collision volume attribute, and use quaternion and matrix transformation (Matrix4x4) to manage the component pose (position and rotation). The tree structure makes the hierarchical relationship of the components clear at a glance, which is convenient for students to understand the structure of the three-electricity system. The attribute binding and pose management enhance the realism of the interaction, so that students can experience the close-to-real operation feeling in the virtual environment.

[0182] Specifically, step S103, initialize the safety rules and process configuration: based on the behavior tree (BehaviorTree) control process, parse the disassembly sequence and safety constraints (such as when disassembling the battery, the constraint condition is "must disconnect high-voltage power first") from the configuration file. By enforcing safety constraints, the students' wrong behaviors in virtual operation are effectively prevented, and their safety awareness and standard operation habits are cultivated.

[0183] It should be further pointed out that in step S1:

[0184] (1) Asynchronous loading technology: using Unity's Addressables or Unreal's Async Loading technology, the asynchronous loading of 3D models (FBX / GLTF format) and scene configuration files (JSON / XML) is realized. Allow resources to be loaded in the background and avoid blocking the main thread, thereby improving the user experience and ensuring that the virtual scene can be started smoothly.

[0185] (2) Real-time update and prompt: during the loading process, the system updates the virtual scene in real time, and prompts the user through the UI about the loading progress or related information, enhancing the user interaction.

[0186] (3) Shader programming and component outline light: using Shader programming technology, the component outline light effect (Outline Shader) is realized, which makes the key components more prominent in complex scenes, facilitating students to identify and operate.

[0187] In this embodiment, as shown in Figure 2 regarding step S2, the user interaction process is:

[0188] Using the sensors built into the VR device (handset, head-mounted display), real-time capture of user hand movements and head rotation in the virtual environment and other operations. Analyze the captured data to identify the user's operation intent, including grabbing, rotating, disassembling or installing components, etc.

[0189] Through the tactile feedback device (vibration motor, force feedback handle) of the VR device, simulate the hand feeling in the real disassembly process, such as the resistance feeling when tightening the screw, the loose feeling when disassembling the component, etc.

[0190] Apply rigid body dynamics principles to simulate physical phenomena such as collision and friction of virtual components during disassembly. When user operations cause components to collide, calculate the motion state and mechanical changes after collision according to the principles of rigid body dynamics, and update the virtual scene in real time.

[0191] According to the preset disassembly sequence and safety constraints, verify whether the user's operation meets the requirements in real time. During the operation process, continuously monitor the state of the virtual components and the user's operation behavior, and timely discover and correct potential safety hazards.

[0192] Use a state machine to manage different states and stages in the virtual disassembly process, such as disassembly state, installation state, and inspection state. When the user performs illegal operations, trigger the state machine rollback mechanism to restore the virtual scene to the state before the operation, and prompt the user to the correct operation method.

[0193] Specifically, step S200, handset / gesture input analysis: the handset detects the component pointed by the user through ray casting, and calculates the interaction point coordinates; the gesture uses the MediaPipe library to identify hand key points, including grabbing gestures and rotating gestures.

[0194] The method of ray casting includes:

[0195] S2000, generation of ray r(t): the handset ray is defined by position O (origin) and direction d (unit vector): t' ≥ 0; where t' is the parameter of the intersection of the ray with each plane, t' ≥ 0 indicates that the ray extends infinitely from the origin along the direction d.

[0196] S2001, collision detection (AABB bounding box intersection test): in order to efficiently detect the intersection of the ray with the components in the scene, an axis-aligned bounding box (AABB) is used for collision detection. The AABB is a cuboid whose boundary is defined by the minimum corner point b min and the maximum corner point b maxDetermine. For the axis-aligned bounding box (AABB) of each component in the scene, let:

[0197] ;

[0198] S2002, for each face of the AABB (in the x, y, and z coordinate axes), calculate the entry parameter t of the ray intersecting with the face enter and exit parameter t exit :

[0199] ;

[0200] Among them, x, y, z represent the coordinate axes, and min represents its minimization parameter;

[0201] S2003, determine whether the ray intersects with the AABB, which needs to meet the intersection conditions:

[0202] ;

[0203] That is, before the ray enters all the faces of the AABB, it must have exited a face of the AABB, so as to ensure that the ray has an intersection with the AABB.

[0204] S2004: When the intersection condition is met, take the nearest legal intersection parameter t hit , calculate the coordinates of the interaction point: ; Then the coordinates of the interaction point ;

[0205] As you can see, the combination of ray casting and AABB bounding box intersection testing enables the controller to accurately detect the component the user is pointing at, improving the accuracy and reliability of interaction. The use of AABB bounding boxes significantly reduces the computational complexity of collision detection, improving the real-time and responsiveness of interaction. Through the controller's ray casting, users can intuitively experience their actions in the virtual scene, enhancing their sense of immersion and experience.

[0206] It should be noted that in step S200, the MediaPipe library is an open source, cross-platform machine learning library that provides efficient, real-time hand key point detection capabilities. The MediaPipe library identifies hand key points and further parses grabbing gestures and rotation gestures. These gestures can correspond to specific operations in the virtual scene, such as grabbing parts, rotating parts, and so on. Gesture input allows users to operate parts in the virtual scene through natural gestures, just like in real life, improving the naturalness and intuitiveness of the interaction. The gesture recognition function of the MediaPipe library can realize the input of multiple gestures, enriching the ways and means of interaction.

[0207] Specifically, in step S201, the tool interacts with the component: detects the collision bounding box (AABB) of the handle and the component, triggers a grabbing event; determines whether the current tool is suitable (such as “whether the torque wrench matches the bolt specification”) through the Drools-Rete algorithm.

[0208] S2010, collision bounding box detection (AABB): based on the handle grabbing point coordinates P hit = (x, y, z) (from the ray projection result), the minimum / maximum corner points of the component AABB bounding box are:

[0209] ;

[0210] By comparing the coordinates of the handle grabbing point and the minimum and maximum corner coordinates of the component AABB bounding box, it is determined whether the handle has entered the bounding box of the component.

[0211] S2011, determine the collision condition as: ; ensure that the handle is in the correct position to interact with the component and avoid misoperation.

[0212] S2012, determine to trigger a grabbing event:

[0213] ;

[0214] S2013, use the Drools-Rete algorithm to perform rule matching to determine whether the current tool is suitable for the target bolt:

[0215] The tool attribute TT of the currently selected tool (three-dimensional model) = (type, min torque , max torque ); wherein, type is the code of the tool model, min torque and max torque are the minimum and maximum torque parameter codes of the tool, respectively;

[0216] The target bolt specification BB in the current working environment = (diameter, required torque ); wherein, diameter is the diameter parameter code of the bolt, and required torque is the torque parameter code required by the bolt;

[0217] Rule matching (Rete network condition evaluation) is a logical expression:

[0218] ;

[0219] Meanwhile, more judgment rules can be added based on the above suitability judgment logic;

[0220] For example, take the insulation of tools as an example: Match 安全 (TT,BB)=Match(T,B)∧(TT.Insulation level ≥ BB.Voltage level);

[0221] After completing this step, the object can be grasped (for example, a tool model, such as Figure 3 shown).

[0222] In S2014, the Rete algorithm is executed: it traverses all nodes, matches them according to tool attributes and bolt specifications, and finds the applicable rules. Finally, the rules are activated: ;

[0223] Where n is the number of nodes, i is the node index; applicable is the rule that is activated;

[0224] As you can see, collision bounding box detection and a grasping event triggering mechanism ensure that the handle interacts with the component in the correct position, preventing misoperation. A tool suitability judgment mechanism ensures that the tool selected by the user matches the actual operation requirements, reducing operational risks. The Rete algorithm quickly finds the appropriate tool, reducing user search time and improving learning efficiency. Simulating real-world disassembly and assembly processes, requiring users to adjust the force and direction of operation based on feedback, cultivates manual skills and muscle reflexes.

[0225] It's important to note that the aforementioned effects are achieved by precisely calculating the positional relationship between the handle's grip point and the component's AABB bounding box, ensuring interaction accuracy. This, combined with the state of the handle's trigger keys, provides users with clear feedback, enhancing the user's immersive experience. The Drools-Rete algorithm is used for rule matching, ensuring the compatibility of tools and bolt specifications, improving operational safety. This efficient node matching mechanism allows for rapid identification of suitable tools, enhancing interaction efficiency and learning outcomes.

[0226] Specifically, such as Figure 4 As shown, step S202, tactile feedback generation: the torque for tightening the screw is calculated according to the physical engine, and the handle vibration motor is driven by the waveform generation algorithm (Sine / PWM).

[0227] Therefore, the implementation of this step involves a tactile feedback generation algorithm, the algorithm process is as follows:

[0228] S2020, mapping of torque to vibration intensity: Assume that the current torque calculated by the physics engine is T (unit: Nm), and define the proportional relationship between the vibration amplitude A and the torque T:

[0229] ;

[0230] where k is a gain coefficient, T max is the maximum torque allowed by the system, which needs to be calibrated according to the handle motor performance; this step is to convert the torque into vibration intensity by defining the proportional relationship between the vibration amplitude A and the torque T according to the current torque T calculated by the physics engine.

[0231] where the gain coefficient k is used to adjust the proportional relationship between the vibration amplitude and the torque, to ensure that the vibration feedback is within a reasonable range. The maximum torque T allowed by the system max needs to be calibrated according to the handle motor performance to prevent motor overload. By mapping the torque to the vibration intensity, users can perceive the change in torque through tactile feedback, enhancing immersion and operation accuracy.

[0232] S2021, generate a sinusoidal vibration signal V sine (t) related to torque using a sinusoidal function:

[0233] where f is the reference frequency, set to the tactile sensitive frequency band (such as 50-200 Hz); and is the initial phase; a nonlinear mapping term A 2 is introduced to enhance high-torque feedback.

[0234] S2022, define the relationship between the duty cycle D ;

[0235] S2023, the PWM waveform V PWM (t) is expressed as: PWM

[0236] ;

[0237] It should be noted that the linear mapping of the duty cycle D and the torque determines the switching time proportion of the PWM waveform.

[0238] S2024, the vibration signal is superimposed and output, and the final driving signal V(t) is mixed with the two waveforms:

[0239] ;

[0240] where a and b are weight coefficients;

[0241] It can be understood that the above PWM waveform changes the proportion of switching time and combines the characteristics of continuous vibration and intermittent vibration by superimposing the two waveforms, making the tactile feedback more rich and realistic. The intermittent vibration feeling of the actual tool under different torques is simulated, enhancing the realism of the tactile feedback.

[0242] ​​Specifically, in practice, by simulating the parameters (thread specifications, torque, equipment types) at the actual three-electric system disassembly component positions, a comparative analysis is performed based on the vibration sensor mounted on the electric screwdriver and the simulated dynamic disturbance in the VR scene. As shown in Figure 8 The scheme of the embodiment can simulate the actual vibration effect (error is about 1-4 mm) more accurately. This shows that the haptic feedback generation algorithm can effectively simulate the torque constraint and vibration feedback in the actual disassembly process, and thus cultivate the students' hand operation ability and muscle reflex.

[0243] It can be understood that by simulating the torque change and vibration feedback in the actual disassembly process, the user can obtain a more realistic operation experience in the VR environment. The haptic feedback enables the user to more accurately perceive the change in torque and the position of the operation, reducing the possibility of misoperation. Through repeated practice and the guidance of haptic feedback, the user can gradually develop correct hand operation habits and muscle reflexes, improving disassembly efficiency and quality.

[0244] Specifically, step S203, as shown in Figure 5 The rigid body dynamics simulation: use the physical engine (Phys) to solve the Newton-Euler equation, calculate the motion of the component, and use the space partition (BVH) to screen the possible collision objects for CPL; mainly used to simulate the disassembly of objects in the scene; it uses the separation axis theorem (SAT) algorithm to detect collision and applies the impulse method (Impulse Method) to calculate the post-collision speed. This series of steps are mainly used to simulate the interaction and collision response of objects in the disassembly scene of the new energy vehicle three-electric system.

[0245] S2030, space partition (BVH) collision screening:

[0246] The Newton-Euler equation is the basic equation of rigid body dynamics, which is used to describe the motion state of a rigid body. By using existing physical engines, these equations can be solved to calculate the motion trajectory of components under the action of external forces, and provide collision detection, constraint solving and other functions to simulate the real physical world. Therefore, the key point is space partition, including the following steps:

[0247] S20300, BVH tree construction, recursively executed for all objects in the scene:

[0248] Calculate a bounding box for each object in the scene to quickly determine the possible collision between objects;

[0249] Divide the object set into left and right sub-trees along the longest axis until the termination condition (the number of objects contained in the leaf node does not exceed k) is met;

[0250] It can be understood that by constructing the BVH tree, the amount of calculation of collision detection can be significantly reduced. Because only the object pairs whose bounding boxes intersect need to be further detected, the efficiency of collision detection is improved.

[0251] S20301, traversing the BVH tree to screen potential collision pairs: for any two nodes N1 and N2, if their bounding boxes intersect, recursively detect their child nodes to screen out CPLs, and the screening method is as follows:

[0252] ;

[0253] Among them, for any two nodes N1 and N2, if their bounding boxes intersect, recursively detect their child nodes to screen out potential collision object pairs (CPL). The screening method recursively or directly judges according to the type of the node (leaf node or non-leaf node);

[0254] S2031, SAT collision detection:

[0255] Projection calculation: for all candidate separation axes a of two objects A and B i (including face normal and cross edge direction), calculate the projection interval Proj A and Proj B :

[0256] ;

[0257] ;

[0258] Among them, q and p are the face normal and the cross edge respectively.

[0259] The separation condition is: if there exists any axis a i satisfying or , then the objects do not collide, otherwise collision occurs.

[0260] It can be understood that this algorithm can accurately determine the collision state between objects and avoid misjudgment and omission.

[0261] S2032, Impulse Method collision response:

[0262] In the normal impulse calculation, let the collision normal vector n, the restitution coefficient e, and the relative velocity v rel = v A -v B , and the impulse J is: ;

[0263] Among them, r A , r B are the vectors from the collision point to the center of mass; v Aand v B are the observed and observed velocities, respectively. Then the post-collision velocity is updated as: ;

[0264] It can be understood that by calculating the impulse and updating the velocity, the motion state of the object after the collision can be truly simulated. During the disassembly process, the collision response between the tool and the component can be accurately simulated, the students' hand operation ability and muscle reflex can be cultivated, and the realism of the operation can be improved.

[0265] Then, the disassembly of the object can be realized, as shown in Figure 6 Through rigid body dynamics simulation and collision detection response, students can observe the physical interaction effect in the disassembly process in real time. The students' sense of participation and interactivity are enhanced. Students can deepen their understanding of the disassembly process of the three-electricity system of new energy vehicles through practice.

[0266] Specifically, as shown in Figure 7 , in step S204, the state machine rollback is triggered: the graph theory is used to detect the disassembly sequence violation and trigger the state machine rollback. First, a graph theory model of the disassembly process is established, then the current state path is tracked, the illegal operation is detected, the rollback target state is determined, and the state stack is updated, while visual feedback is provided to prompt the user.

[0267] In S2040, the graph theory modeling of the disassembly process is performed: let the disassembly process be a directed graph G=(V,E), where the vertex set V={v1,v2,...,v n} represents the legal operation state (such as "power off", "bolt disassembled"); the edge set E V i ×V j represents the allowed state transition (such as );

[0268] That is, each vertex represents a specific operation state, which is a legal node in the disassembly process. The edge represents the legal transition from state v i to state v j .

[0269] In S2041, the current state path tracking is performed: define the state history stack S=[s0,s1,...,s t'' ], where s K ∈V represents the state of the Kth step, and the current state is s t'' ;

[0270] In S2042, when the user performs an action a, the state transition is triggered, and if the transition is illegal: (s t ,s'')∉E, the rollback condition RollbackTrigger is triggered;

[0271] S2043, rollback target state determination: find the nearest legal predecessor state s by reverse search of the history stack S p , such that ∃(s p , s valid ) ∈ E, where s valid is the legal state of the current scenario.

[0272] S2044, update the state after rollback as S ← [s0, s1,..., s p ];

[0273] Preferably, simulate a red flashing special effect and bind the rollback target state, as shown in Figure 7 , to prompt the user of the current intervention point.

[0274] It can be understood that through graph modeling and path tracking, the system can accurately determine the legality of user operations and ensure that the disassembly sequence meets the actual requirements. Avoid disassembly failure or damage caused by incorrect operation sequence, improve the accuracy and reliability of teaching. When the user attempts to perform illegal operations, the system can detect and trigger rollback immediately, restoring the state to the nearest legal predecessor state. Provide immediate feedback to help users quickly correct incorrect operations and improve learning efficiency and operation accuracy.

[0275] Prompt the user of the current intervention point through a red flashing special effect, enhance the user's visual perception and operation guidance. Make the user more clearly understand the specific location of the incorrect operation, improve the intuitiveness and interactivity of teaching, and cultivate the students' hand operation ability and muscle reflex. And, by reverse search of the history stack to determine the rollback target state, the system can flexibly cope with various illegal operation situations.

[0276] In this embodiment, regarding step S3, disassembly process logic control: define disassembly steps based on directed acyclic graph (DAG), use breadth-first search (BFS) to push the current executable operation (such as "the next step should disconnect the high-pressure interface"); At the same time, through the rule engine, monitor illegal operations in real time and provide an alarm;

[0277] Use a hash table (Hash Table) to map the association between tools and actions (such as "electric wrench → tighten bolt"), and match user intent through cosine similarity.

[0278] It is noted that in this step, the DAG structure is a graph structure in which the nodes represent disassembly steps, the directed edges represent the dependency relationship between steps, and there is no loop. The disassembly process of the new energy vehicle three-electric system is divided into a series of steps, each step is taken as a node of the DAG, and the order and dependency relationship between the steps are represented by the directed edges. The DAG ensures the order and dependency of the disassembly steps, avoiding the occurrence of circular dependency and deadlock. In the DAG, starting from the current step node, the BFS algorithm is used to traverse the adjacent nodes to find all currently executable operations (i.e. steps without completed preconditions). The BFS ensures that the executable operations are pushed in the hierarchical order of the steps, guiding the user to disassemble in the correct order.

[0279] The hash table is a data structure for fast lookup and mapping of key-value pairs. The tool (such as an electric wrench) is taken as the key, and the possible actions (such as screwing bolts) are taken as the value, which are stored in the hash table. The hash table provides fast tool-action mapping, supports real-time response to user operations, and improves the smoothness of interaction.

[0280] The cosine similarity is an index for measuring the similarity between two vectors, ranging from [-1, 1], with 1 indicating complete similarity. The user operation (such as gesture, voice command) is converted into vector representation, and the cosine similarity is calculated with the predefined operation vector. The cosine similarity matches the user's intention, allowing the system to understand and respond to the user's ambiguous or incomplete operations, improving the naturalness and flexibility of interaction.

[0281] Specifically, in step S300, the disassembly step pushing based on DAG: the algorithm of this step is defined based on directed acyclic graph (DAG).

[0282] Let the disassembly flowchart be G=(V,E), where the vertex set V={v1,v2,...,v m} represents a total of m operation steps v i (such as v i ="power off", v j ="remove the shell"); the edge set E V×V represents the step dependency relationship (such as E(v i ,v j )∈E represents that "power off" must be performed before "remove the shell");

[0283] The characteristics of the directed acyclic graph (DAG) are as follows:

[0284] (1) Directionality: the edge E(v i ,v j ) represents the directionality from vertex vi to vertex vj, i.e. operation v i is the precondition of operation v j .

[0285] (2) Acyclic: There are no loops in the graph, i.e. there is no sequence of vertices v1, v2,..., vn such that E(v1, v2), E(v2, v3),..., E(vn-1, vn), E(vn, v1) all hold simultaneously. This guarantees the order of steps and avoids deadlocks or circular dependencies. k k

[0286] The flow of the DAG-based disassembly step pushing algorithm is as follows:

[0287] 1) Initialization: Determine the current executable operation set, usually starting from the vertices with no pre-dependencies (i.e. vertices with in-degree 0).

[0288] 2) Step pushing:

[0289] 2.1) Traverse the current executable operation set and present the operations to the user.

[0290] 2.2) The user selects an operation and executes it.

[0291] 2.3) Update the graph G, mark the executed operation vertices and remove the corresponding edges.

[0292] 2.4) Recalculate the current executable operation set, i.e. find all unexecuted vertices with in-degree 0.

[0293] 3) Repeat: Repeat the above steps until all operations are executed.

[0294] Use an adjacency list or adjacency matrix to represent the graph G, which facilitates fast lookup of adjacent vertices and updating of edge information. Maintain an in-degree array to record the in-degree value of each vertex, which facilitates fast lookup of vertices with in-degree 0. After the user executes an operation, dynamically update the graph G and the in-degree array to ensure the efficiency of the algorithm.

[0295] ​​It can be understood that by defining the disassembly steps through the DAG, the order and dependency of the steps are ensured, and the operation confusion and errors are avoided. The directionality and acyclicity of the DAG guarantee the logical order between the steps, avoiding the occurrence of circular dependency and deadlock situation. The algorithm ensures that the user can only perform the current executable operation by dynamically updating the graph G and the in-degree array. The DAG-based disassembly step pushing algorithm can automatically push the current executable operation, improving the teaching efficiency and accuracy. Through the initialization, step pushing and repeated execution steps, the algorithm automatically guides the user to perform the disassembly operation in the correct order, reducing human errors and omissions. By simulating real disassembly steps and dependency relationships, the students' hand operation ability and muscle reflex are cultivated. The order and accuracy of the disassembly steps are ensured, so that students can obtain an experience close to real operation in a virtual environment. Students need to perform operations in the correct order, which helps to cultivate their hand operation ability and muscle reflex.

[0296] Specifically, step S301, current executable step set generation (BFS traversal): define the adjacency matrix A, where A ij =1 if and only if E(v i ,v j )∈E. Starting from the current completed step set S done , find all directly reachable steps through BFS:

[0297] ;

[0298] User interface prompts executable operation ;

[0299] Where the priority order is preset based on the safety level or topological order.

[0300] The adjacency matrix A is a two-dimensional matrix, which is used to represent the dependency relationship between disassembly steps. The elements A ij in the matrix represent the dependency relationship from step v i to step v j . When A ij =1, it means that there is an edge E(v i ,v j ), that is, step v i must be completed before step v j ; when A ij =0, it means that there is no such dependency relationship. Through the adjacency matrix A, the precedence relationship between disassembly steps can be intuitively represented, providing a basis for subsequent BFS traversal.

[0301] Breadth-First Search (BFS) is used to traverse or search the data structure of a graph. It starts from the starting node, first visits all adjacent nodes, then visits all unvisited adjacent nodes of these adjacent nodes in turn, until all reachable nodes in the graph are visited. From the current completed step set S done , the adjacency matrix A is traversed by BFS to find all directly connected steps with S done and not yet completed, i.e. the directly reachable step set S next .

[0302] Specifically, for each step vi in Sd one , check all its adjacent steps v j (A ij =1), if v j is not visited (i.e. not in S done ) and vj has no other predecessor steps (i.e. there is no (v k ,v j )∈E and v k is not in S done ), then v j is added to S next . By traversing the adjacency matrix A by BFS to generate the directly reachable step set S next , it can be ensured that students can only perform the operations allowed in the current stage, avoiding safety problems caused by misoperation, and also meeting the sequence requirements of the disassembly process.

[0303] After determining the directly reachable step set S next , an optimal step needs to be selected as the next executable operation NextStep. This step uses priority (Priority) as the selection criterion. Priority is preset based on safety level or topological order. For each step v next in S j , calculate its priority Priority(v j ). Then, select the step with the highest priority as NextStep. This can be achieved by a simple comparison operation, i.e. find v j that makes Priority(v j ) maximum. By introducing priority and selecting the step with the highest priority as NextStep, students can be guided to operate in a safer and more reasonable order. For example, steps with high safety level may involve high-voltage operations and need to be completed first; topological order ensures that the dependency relationship between steps is correctly maintained. This mechanism helps to cultivate students' safety awareness and operation standards.

[0304] Specifically, step S302, rule engine violation monitoring: in the virtual disassembly teaching process, each operation will trigger a change in system state. Let the current operation be a t" , which corresponds to the state transition from the previous state s t"-1 to the current state s t" . The system needs to track and record each operation and the state transition it triggers in real time for subsequent legality verification:

[0305] Let the current operation be a t" , the state transition , and the legality verification function AlarmTrigger:

[0306] ;

[0307] Then, when the alarm is triggered, it is forced to roll back to the nearest legal state (see the state machine rollback algorithm mentioned earlier).

[0308] In other words, the logic of this step is:

[0309] Condition one: , that is, check whether the state transition from the previous state s t-1 to the current state s t exists in the predefined edge set E. If it does not exist, it means that the state transition is illegal.

[0310] Condition two: tool matching failure, that is, check whether the tool used by the current operation conforms to the preset tool matching rule. If it does not conform, it means that the tool is used improperly and is illegal.

[0311] When the AlarmTrigger function returns 1, it means that the current operation is illegal; when it returns 0, it means that the current operation is legal. The system will trigger an alarm to remind the student that the current operation has an error. At the same time, in order to correct the error and restore to a legal state, the system will perform a state rollback operation. The state rollback operation relies on the state machine rollback algorithm mentioned earlier. This algorithm will roll back the system state to the nearest legal state s_{t-n} according to the historical state record, where n is the number of steps of rollback. During the rollback process, the system will cancel all illegal state transitions and tool usage records to ensure that the system returns to a legal and consistent state.

[0312] Specifically, step S303, tool-action mapping and intent matching hash table stores the tool-action relationship, defines a hash function H: KK→V, where the key KK is the tool type (such as "electric wrench") value, and V is the allowed action set, such as {screw bolt, remove nut};

[0313] Through the hash function, the allowed action set corresponding to the given tool type can be quickly found. The hash table allows lookup with O(1) time complexity, so that the action set allowed by the given tool type can be quickly determined. New tool types and corresponding action sets can be easily added without modifying existing logic.

[0314] Match user intent based on cosine similarity: Let the feature vector of the user's current operation be u = (u1, u2,..., u m ) (including torque and angle sensor data), and the target action requirement vector be v a = V = (v1, v2,..., v m ). The matching degree is ; where ||·|| is the logical operator "or".

[0315] Select the optimal action ;

[0316] Through cosine similarity calculation, the user's current operation intent can be accurately identified, and the action that best meets the requirements can be selected. Automatically matching user intent and suggesting the optimal action can reduce user operation steps and improve interaction efficiency. In the process of repeated matching and selection of the optimal action, students can gradually master the correct disassembly skills and force control methods, and thus develop their hand operation ability and muscle reflex. This mechanism helps students more accurately perform disassembly operations in future actual work and quickly adapt to different tools and environments.

[0317] In this embodiment, regarding step S4, teaching evaluation and feedback: record the timestamp, operation type and component ID information, and store it as a time series database (InfluxDB).

[0318] Based on the weighted method as the scoring rule, randomly inject faults (such as "communication interruption of the electric control system"), and use decision trees (ID3 / C4.5) to determine whether the student's troubleshooting path is correct.

[0319] Specifically, step S400, operation data collection and time series storage: when the user performs key operations, including grabbing components, using tools and completing steps, data recording is triggered; the recorded fields include:

[0320] (1) Timestamp: accurate to milliseconds (such as 2024-05-20T14:23:45.678Z);

[0321] (2) Operation type: classification label (such as disassembling the battery shell, incorrect non-insulated operation);

[0322] (3) Component ID: unique identifier (such as a certain block of battery marked as Battery.Module3.Cell25);

[0323] (4) Additional data: operating parameters, tool model, and time spent.

[0324] The InfluxDB stores information including:

[0325] (1) Data Sharding: Partitioned by student ID or operation stage, accelerating queries.

[0326] (2) Data Compression: Delta Encoding for repeated operations (mainly for continuous screwing).

[0327] (3) Tag Indexing: Inverted index for high-frequency query fields (such as component ID, error type).

[0328] Specifically, step S401, data stream pipeline execution: real-time stream processing is performed through Kafka or MQTT to transmit data from the VR client to the database; invalid operations (such as false triggering caused by handle shaking) are filtered out.

[0329] The method is that: the VR client sends the collected operation data to the message queue system in real time. The message queue system performs deduplication processing on the data, filtering out repeated or invalid operation data. The verification operation ensures that the format and content of the data meet the expected requirements. The processed data is finally stored in the time series database.

[0330] Specifically, step S402, diagnosis logic of fault simulation: based on predefined common faults (such as communication interruption of electronic control, short circuit of battery monomer), randomly inject faults after the user performs specific operations (such as disassembling the electronic control shell); trigger faults every fixed time interval (such as 5 minutes) with a probability p; and simulate virtual component abnormalities (such as instrument panel displaying error codes, component smoke special effects).

[0331] Specifically, step S403, evaluation of student troubleshooting path (decision tree logic): the order of student troubleshooting steps (such as checking the power supply first, then testing the signal line); whether to use the correct detection device (such as using a multimeter to measure voltage); the response time is the delay from the occurrence of the fault to the start of troubleshooting;

[0332] The method is that: the student's troubleshooting path and the corresponding "correct / incorrect" label are extracted from historical data. Use ID3 / C4.5 algorithms to train a decision tree model, and select the feature with the highest information gain as the basis for judgment. Real-time acquisition of student's troubleshooting steps and order, detection devices used, etc. Compare the student's operation with the optimal path generated by the decision tree to evaluate the student's troubleshooting efficiency and accuracy. If the student continuously deviates from the correct path, trigger a prompt (such as "suggest checking CAN bus connection"). Based on the path matching degree and troubleshooting efficiency, calculate the score based on the weighted method as the final evaluation result of the student.

[0333] Specifically, in step S403, decision tree training (ID3 / C4.5): select the feature with the highest information gain (such as "whether to detect the master module first") from the "correct / incorrect" troubleshooting path marked in the historical data; compare the student's operation with the optimal path generated by the decision tree in real time. If the student continuously deviates from the correct path, trigger a prompt (such as "suggest checking the CAN bus connection"); based on the path matching degree and troubleshooting efficiency, calculate the score based on the weighted method.

[0334] Embodiment Two: Based on Embodiment One, this embodiment further provides a preferred scheme for step S201: in step S2021, by combining the interactive point coordinates P hit (Previous calculation results) Adjust the vibration space attenuation (the farther the distance, the weaker the vibration). The scheme is as follows:

[0335] S20210, let the interactive point coordinates P hit =(x hit ,y hit ,z hit ), the current handle position coordinates P handle =(x handle ,y handle ,z handle ), then the distance d between the two points can be calculated by the three-dimensional space distance formula: ;

[0336] S20211, in order to map the distance d to the vibration intensity A, use an inverse proportional attenuation function, where the vibration intensity A is proportional to the inverse of the distance d, while considering a basic vibration intensity A0 and an attenuation coefficient k decay : ;

[0337] Where δ is a positive number used to control the speed of attenuation. When δ is larger, the vibration intensity decays faster with the increase of distance. Avoid A approaching infinity when d approaches 0.

[0338] It can be understood that in actual operation, when the tool (such as a screwdriver) is far away from the component that needs to be operated, the vibration felt by the operator will usually weaken due to the increase of the force arm and the decrease of the transmission efficiency. By simulating this vibration attenuation phenomenon in the VR environment, the virtual operation can be made closer to the real experience, enhancing the user's sense of immersion and the sense of reality of the operation.

[0339] Vibration feedback is an important basis for operators to judge the contact state and operating force of the tool and the component. When the tool approaches or contacts the component, the enhancement of vibration feedback can help the operator to more accurately locate the operating position and control the operating force. Conversely, when the tool is away from the component, the weakening of vibration feedback can avoid misleading the operator and improve the accuracy of operation. By simulating the real vibration attenuation phenomenon, the operator can be guided to develop correct operation habits in the VR environment. The operator needs to learn to judge the relative position and operating state of the tool and the component according to the change of vibration feedback, so as to be more skilled in actual operation.

[0340] Embodiment three: on the basis of step S300 of embodiment one, a series of reference-dependent relationship examples are further provided:

[0341] Let V be a set of disassembly steps, where each element v i represents a specific disassembly step. For example:

[0342] v1: disconnect the high-voltage connection of the battery; v2: disassemble the battery pack shell; v3: remove the battery module; v4: disconnect the motor power line; v5: disassemble the motor fixing bolt; v6: remove the motor; v7: disconnect the control system connection; v8: disassemble the control module shell; v9: remove the control module;

[0343] The edge set E describes the dependency relationship between these steps. For example:

[0344] E(v1,v2): indicates that before disassembling the battery pack shell (v2), the high-voltage connection of the battery (v1) must be disconnected. E(v1,v3): indicates that before removing the battery module (v3), the high-voltage connection of the battery (v1) must also be disconnected. E(v4,v5): indicates that before disassembling the motor fixing bolt (v5), the motor power line (v4) needs to be disconnected. E(v5,v6): indicates that before removing the motor (v6), the motor fixing bolt (v5) must be disassembled. E(v7,v8): indicates that before disassembling the control module shell (v8), the control system connection (v7) needs to be disconnected. E(v8,v9): indicates that before removing the control module (v9), the control module shell (v8) must be disassembled.

[0345] These dependency relationships can be drawn into a directed graph, where each node represents a disassembly step and each directed edge represents a dependency relationship. For example:

[0346] v1→v2, v1→v3, v4→v5, v5→v6, v7→v8, v8→v9;

[0347] According to this dependency graph, the planned disassembly sequence is as follows:

[0348] Disconnecting the high-voltage connection of the battery (v1), disassembling the battery pack shell (v2), taking out the battery module (v3), disconnecting the motor power line (v4), disassembling the motor fixing bolt (v5), taking out the motor (v6), disconnecting the electric control system connection (v7), disassembling the electric control module shell (v8), taking out the electric control module (v9);

[0349] The above all embodiments only express the relevant actual application implementation of the present application, the description is more specific and detailed, but can not therefore be understood as the restriction of the patent scope of the invention. It should be pointed out that for the ordinary skilled person in the art, without departing from the concept of the present application, can also make several deformation and improvement, these all belong to the protection scope of the present application. Therefore, the patent protection scope of the present application should be limited by the appended claims.

[0350] For those skilled in the art, it can be further realized that the units and algorithm steps of each example described in combination with the embodiments disclosed in the present application can be realized by electronic hardware, computer software or combination of both. In order to clearly show the interchangeability of hardware and software, the components and steps of each example have been described in the above description. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0351] Meanwhile, those skilled in the art can understand that all or part of the processes in the above-mentioned all embodiments can be completed by a computer program instructing related hardware, and the computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, the computer program can include the processes of the above-mentioned embodiments. Any reference to memory, storage, database or other medium provided by the present application and used in the embodiments can include non-volatile and / or volatile memory. The non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. The volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM) and memory bus dynamic RAM (RDRAM).

Claims

1. A VR interaction method for virtual disassembly and assembly teaching of a three-electricity system of a new energy vehicle, characterized in that, Comprise the following steps: S1, load 3D model and scene configuration file, real-time update virtual scene, build scene topology and record component relationship; S2, capture and analyze user's operation intention in VR environment, generate haptic feedback simulation real disassembly hand feeling, through rigid body dynamics simulation collision response, and trigger state machine rollback to correct illegal operation; The execution method of the S2 comprises: S200, the handle detects the component pointed by the user through ray casting, and calculates the interaction point coordinates;Gesture uses MediaPipe library to identify hand key points, including grabbing gesture and rotating gesture; The execution method of the S200 comprises: The handle ray is defined by a position O as the origin and a direction d as the unit vector: t' ≥ 0; where t' is the parameter at which the ray intersects the six faces of the axis-aligned bounding box. S2001, for the axis-aligned bounding box of each component in the scene, set: ； S2002, calculate the entering parameter t of the intersection of the ray with each plane enter and the exiting parameter t exit ; Wherein, x, y, z represents the coordinate axis, min represents its minimum parameter; S2003, calculate intersection condition: ; S2004, take the nearest legal intersection parameter t hit , calculate intersection point coordinates: ； then the intersection point coordinates ; S201, detect the collision bounding box of the handle and the component, trigger the grabbing event;Determine whether the current tool is suitable through Drools-Rete algorithm; The execution method of the S201 comprises: S2010, based on the interaction point coordinates P hit = (x, y, z), calculate the minimum / maximum corner points b of the component AABB min , b max ; S2011, determining the collision condition as ; S2012, determine the grabbing event GrabEvent; S2013, current tool attribute TT = (type, min torque , max torque ) ; wherein type is the code of the tool model, min torque and max torque are the minimum and maximum torque parameter codes of the tool respectively; target bolt specification BB = (diameter, required torque ) ; wherein diameter is the diameter parameter code of the bolt, required torque is the required torque parameter code of the bolt; The rule matching is a logical expression: ； S2014, Drools-Rete algorithm matches tool type through node and associates bolt demand, finally activates the rule; S202, calculate the torque of screwing according to the physical engine, and drive the handle vibration motor through the waveform generation algorithm; S2020, set the current torque calculated by the physical engine as T, define the proportional relationship between the vibration amplitude A and the torque T; ; k is a gain coefficient, T max is the maximum torque allowed by the system; S2021, generate a sine wave vibration signal V related to the torque sine (t): ； Wherein, f is the reference frequency;Phi is the initial phase; S2022, PWM (pulse width modulation) vibration signal generation: Defining the relationship of the duty cycle D e [0, 1] to the torque ; S2023, PWM waveform V PWM (t) is within a period T PWM is: ； S2024, vibration signal superposition and output, finally drive signal V(t) mixed two waveforms: ； Wherein, alpha and beta are weight coefficients; S203, use physical engine to solve Newton-Euler equation, calculate component motion, use space division to screen possible collision objects CPL;Adopt separation axis theorem algorithm to detect collision, and apply impulse method to calculate post-collision velocity; S204, detect disassembly sequence violation through graph theory, trigger state machine rollback; S3, define disassembly steps based on directed acyclic graph, use breadth-first search to push current executable operation;Real-time monitoring of illegal operation is carried out through rule engine;Use hash table to map the association between tool and action, and match user intention through cosine similarity; S4, record timestamp, operation type and component ID information, and store it as time series database;Based on weighted method as scoring rule, randomly inject fault, and use decision tree to judge whether the student's troubleshooting path is correct.

2. The VR interaction method of claim 1, wherein: The execution steps of the S1 comprise: S100, load 3D model and scene configuration file;Real-time update virtual scene and prompt information, generate floating UI in VR screen space;Use multiple camera viewports to support VR split screen display; S101, start model LOD hierarchical management, so that the model is only rendered in front of the relative camera viewport; S102, record the parent-child hierarchical relationship of components through tree data structure, bind quality attribute and collision volume attribute, and use quaternion and matrix transformation to manage component pose; S103, based on the constructed behavior tree control flow, parse disassembly sequence and safety constraints from configuration file.

3. The VR interaction method of claim 1, wherein: The execution method of the S203 comprises: S2030, space division screen collision pair; BVH tree for all objects in the scene recursively: Calculate the bounding box; Split the object set along the longest axis into left and right sub-trees; The termination condition is that the leaf node contains ≤k' objects; Then traverse the BVH tree to screen potential collision pairs. For any two nodes N1, N2, if their bounding boxes intersect, recursively detect their child nodes to screen out CPLs. S2031, for all candidate separation axes a of the two objects A, B i , compute the projection interval Proj A and Proj B ; wherein the separation condition is: if there is any axis a i satisfies or is then the object does not collide, otherwise a collision occurs; S2032, collision response based on impulse method, update the speed after collision; The execution method of S204 includes: S2040, set disassembly flow is directed graph G=(V, E), wherein vertex set V={v1, v2,..., v n} represents a legal operating state; edge set E V i ×V j represents the allowed state transition; S2041, define state history stack S = [s0, s1,..., s t'' K], where s K ∈ V represents the state of the Kth step, the current state is s t'' K-1 S2042, when the user performs action a then trigger state transition , if the transition is not legal: (s t , s'') ∉ E, then trigger rollback condition RollbackTrigger.

4. The VR interaction method of claim 1, wherein: In S3, including: S300, set disassembly flowchart G=(V, E), wherein vertex set V={v1, v2,..., v m} represents a total of m operation steps v i ; edge set E VxV represents step dependency; S301, define the adjacency matrix A, where A ij =1 if and only if E(v i ,v j )∈E; from the current set of completed steps S done , find all directly reachable steps by BFS: ； Executable operations prompted by user interface ; Wherein, the priority order is based on the security level or topological order to be preset; S302, rule engine violation monitoring: Let the current operation be a t" corresponding state transition , the check legality function AlarmTrigger: ； Then, when the alarm is triggered, force rollback to the nearest legal state; S303, hash table storage tool-action relationship, define hash function H: KK→V, where the key KK is the tool type value, and V is the vertex set.

5. The VR interaction method of claim 1, wherein: In S4, based on pre-defined faults, randomly inject faults after user operation; trigger faults every fixed time with probability p; And simulate virtual component abnormalities; The contents of the check include: The order of the student troubleshooting steps; Whether to use the correct detection equipment; Response time is the delay from fault occurrence to start troubleshooting; Select the feature with the highest information gain from the "correct / incorrect" troubleshooting path marked in the historical data; Real-time comparison of student operations and the optimal path generated by the decision tree; If the student continuously deviates from the correct path, trigger a prompt; Based on the path matching degree and troubleshooting efficiency, calculate the score based on the weighted method.

6. A VR interaction system for virtual disassembly and assembly teaching of a new energy vehicle three-electric system, characterized in that: The system includes a processor, a memory connected to the processor, and program instructions stored in the memory, which when executed by the processor, cause the processor to perform the VR interaction method of any one of claims 1-5.

7. The VR interaction system of claim 6, wherein, Including: The processor is connected with a three-dimensional model and scene management module, a physical simulation and interaction module, a disassembly process logic control module, a teaching and evaluation module and a VR device.

Images