VR interaction method and system for virtual disassembly and assembly teaching of three-electric system of new energy automobile

Abstract

The invention discloses a VR interaction method and system for virtual disassembly and assembly teaching of a three-electric system of a new energy automobile. The invention relates to the technical field of VR teaching. And loading a 3D model and a scene configuration file, updating a virtual scene in real time, constructing a scene topological structure, recording a component relationship, and initializing security rules and process configuration. Loading the 3D model and the scene configuration file by using an asynchronous loading technology based on a GPU (Graphics Processing Unit); updating the virtual scene in real time and prompting information; the father-son hierarchical relation of the components is recorded through the tree-shaped data structure, the disassembly and assembly process is controlled based on the constructed behavior tree, and it is ensured that students operate according to the correct sequence and steps. Safety rules and flow configuration are built in, for example, high-voltage electricity must be cut off firstly when the battery is disassembled and assembled, and safety accidents caused by misoperation are effectively avoided. And when the operation of the student is illegal, the state machine is triggered to roll back to the latest legal state, so that the student is helped to correct errors in time, and the smooth operation of the disassembly and assembly process is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of VR teaching, specifically to the direction of automobile disassembly and assembly teaching and VR technology, and particularly to a VR interaction method and system for virtual disassembly and assembly teaching of the three-electric systems of new energy vehicles. Background Art

[0002] Virtual Reality (VR) technology is an interactive technology that generates a three-dimensional dynamic simulation environment through a computer to provide users with an immersive experience. VR technology can simulate the real scene of a new energy vehicle workshop, including vehicles, tool vehicles, parts vehicles, protective equipment, etc. Students can walk freely in the virtual scene, observe the overall vehicle structure and each component, and interact with the scene through a handle to bring an immersive disassembly and assembly experience. At the same time, students can select any task on the tablet for practical operation training and self-assessment, and operate according to the standard operation steps.

[0003] However, there are some defects in the existing technologies: CN108711326A discloses a virtual simulation-based automobile education and training system (publication date: October 26, 2018); CN119477615A discloses a virtual training method for automobile vocational education (publication date: February 18, 2025); CN118197120A discloses a new energy vehicle disassembly process teaching demonstration system based on VR interaction technology (publication date: June 14, 2024); these existing technologies are all simply building VR demonstration scenes and demonstrating automobile disassembly and assembly teaching according to a fixed demonstration process. Students cannot participate in them and can only passively understand the visualized knowledge, and their actual interaction effect is not strong and the teaching effect is not ideal.

[0004] CN113744590A discloses a VR interactive teaching device for virtual disassembly, assembly and detection of the high-voltage part of a pure electric vehicle (publication date: December 3, 2021); CN108806385A discloses a VR immersive multi-person collaborative automobile teaching and training system and teaching method (publication date: November 13, 2018). Although these existing technologies can combine VR demonstration with disassembly and assembly interaction, enabling students to interact with the VR scene and simulate the disassembly and assembly of automobile components, the structure of the three-electric systems of new energy vehicles is complex and there are many components. In particular, some specific mechanisms must strictly follow specific disassembly processes (such as "battery pack → module → battery cell"); however, the above traditional technologies do not add necessary constraint items, resulting in students not being able to learn strict disassembly and assembly processes.

[0005] This defect of weak constraint effect further leads to another technical problem: the removal of screws in the three-electric system of new energy vehicles must strictly follow a specific torque, because in the battery system, inappropriate torque may cause the battery connection to loosen or be damaged, affecting the performance and life of the battery. For example, when removing the battery connection bolts, excessive torque may damage the battery connector, resulting in poor contact or short circuit. The bolt connection in the motor and electronic control system also requires precise torque control. Inappropriate torque may cause displacement of internal parts of the motor, increased wear, or poor connection of the electronic control system, affecting the performance and reliability of the entire three-electric system. Traditional technology can neither achieve the corresponding conditional constraints, nor simulate the tactile feel brought about by the selection and actual application of torque, nor can it cultivate students' applicable force size (muscle reflex) under various torque conditions, and cannot effectively cultivate students' practical experience.

[0006] To this end, the present invention proposes a VR interactive method and system for virtual disassembly and assembly teaching of the three-electric system of new energy vehicles. Summary of the invention

[0007] In view of this, the present invention hopes to provide a VR interactive method and system for virtual disassembly and assembly teaching of the three-electric system of new energy vehicles to solve or alleviate the technical problems existing in the prior art, namely: how to further provide necessary constraint functions for each process demonstration item, while giving full play to the purpose of process teaching, and further introduce a disassembly and assembly torque simulation mechanism to simulate the torque constraint required for disassembly and assembly of the three-electric system in practice, thereby cultivating students' manual operation ability and muscle reflex, and at least providing a beneficial choice for this; the technical solution of the present invention is implemented as follows: First, the VR interactive method for virtual disassembly and assembly teaching of the three-electric system of new energy vehicles: The present invention aims to build a new energy vehicle education and training system through virtual reality (VR) technology to achieve immersive disassembly and assembly experience and teaching evaluation feedback. The solution includes four core parts: resource loading and scene construction, user interaction processing, disassembly and assembly 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 achieve interaction with virtual components and simulate the real disassembly and assembly feel. The disassembly and assembly steps are defined based on a directed acyclic graph to ensure accurate process logic and real-time monitoring of illegal operations. At the same time, student operation data is recorded to provide intelligent evaluation feedback and fault simulation troubleshooting training. This solution effectively solves the problems of weak interactivity and insufficient constraints on disassembly and assembly processes in traditional automotive education and training, and improves the teaching effect and students' learning experience.

[0008] (II) Technical solution: 2.1 Step S1, system initialization and scene loading: Load the 3D model and scene configuration file, update the virtual scene in real time, construct the scene topology structure and record the component relationships, and initialize the security rules and process configurations.

[0009] 2.1.1 Step S100, Resource Loading and Model Instantiation: Use GPU-based asynchronous loading technology to load the 3D model and scene configuration file; update the virtual scene in real time and prompt information, implement component contour lighting through Shader programming, and generate floating UIs in the VR screen space; use multiple camera viewports to support VR split-screen display.

[0010] 2.1.2 Step S101, Start Model LOD Hierarchical Management: Render the model only in front (close range) relative to the camera viewport to reduce GPU load; 2.1.3 Step S102, Scene Topology Structure Construction: Record the parent-child hierarchical relationships of components through a tree-like data structure (taking the power supply system of a car as an example, its parent-child hierarchical relationships are "battery pack → module → cell"), and bind the mass attribute and collision volume attribute, and use quaternions and matrix transformations to manage the component pose (Position&Rotation).

[0011] 2.1.4 Step S103, Initialize Security Rules and Process Configurations: Based on the constructed behavior tree to control the process, parse the disassembly and assembly sequence and safety constraints from the configuration file (for example, when disassembling and assembling the battery, the constraint condition is "the high voltage must be disconnected first").

[0012] 2.2 Step S2, User Interaction Processing: Capture and parse the user's operation intention in the VR environment, generate tactile feedback to simulate the real disassembly and assembly feel, simulate the collision response through rigid body dynamics, ensure the correctness and safety of the disassembly and assembly operations, and trigger the state machine rollback to correct illegal operations.

[0013] 2.2.1 Step S200, Handle / Gesture Input Parsing: The handle detects the component pointed by the user through ray casting and calculates the interaction point coordinates; gestures use the MediaPipe library to identify hand key points, including grasping gestures and rotating gestures.

[0014] The methods of ray casting include: S2000, Generation of Ray r(t): The handle ray is defined by the position O (origin) and direction d (unit vector): r(t)=O+t⋅d, t≥0; where t is the parameter at which the ray intersects each plane; S2001, Collision Detection (AABB Bounding Box Intersection Test): For the axis-aligned bounding box (AABB) of each component in the scene, set: ; S2002, Calculate the parameter t of the intersection of the ray with each plane enter and t exit : ; where x, y, z represent the coordinate axes, and min represents its minimization parameter; S2003, Calculate the intersection condition: ; S2004, Take the nearest legal intersection point parameter t hit , and calculate the interaction point coordinates: ; Then the interaction point coordinates ; 2.2.2 Step S201, Tool-Component Interaction: Detect the collision bounding box (AABB) of the handle and the component, and trigger the grasping event; use the Drools-Rete algorithm to determine whether the current tool is applicable (such as "whether the torque wrench matches the bolt specification").

[0015] S2010, Collision Bounding Box Detection: Based on the handle grasping point coordinates P hit =(x, y, z) (from the ray casting result), the minimum / maximum corner points b min and b max of the component AABB bounding box are: ; S2011, Determine the collision condition as: ; S2012, Determine to trigger the grasping event GrabEvent: ; S2013, Tool Applicability Judgment (Drools-Rete Algorithm): The current tool attribute TT = (type, min torque , max torque ); where type is the encoding of the tool model, and min torque and max torque are the minimum and maximum torque parameter encodings of the tool respectively; The target bolt specification BB = (diameter, required torque ); where diameter is the diameter parameter encoding of the bolt, and required torqueIt is the torque parameter encoding required for the bolt; Rule matching (Rete network condition evaluation), which is a logical expression: ; S2014, the Rete algorithm is executed: By matching the tool type at the node and associating the bolt requirements, the rule is finally activated: ; Among them, n is the number of nodes, and i is the node index; applicable is the activated rule; 2.2.3 Step S202, haptic feedback generation: Calculate the torque of screwing the screw according to the physical engine, and drive the handle vibration motor through the waveform generation algorithm (Sine / PWM).

[0016] The implementation of this step involves a haptic feedback generation algorithm, and the process is as follows: S2020, mapping from torque to vibration intensity: Let the current torque calculated by the physical engine be T (unit: Nm), and define the proportional relationship between the vibration amplitude A and the torque T ; Among them, k is the gain coefficient, and T max is the maximum torque allowed by the system, which needs to be calibrated according to the performance of the handle motor; S2021, generate a sine wave vibration signal V sine (t): ; Among them, f is the reference frequency, set to the haptic sensitive frequency band (such as 50–200 Hz); ϕ is the initial phase; a non-linear mapping term A∝T can be introduced 2 to enhance the high-torque feedback.

[0017] The vibration spatial attenuation can be adjusted in combination with the interaction point coordinates P hit (the calculation result above) (the vibration becomes weaker as the distance increases).

[0018] S2022, PWM (pulse width modulation) vibration signal generation: Define the relationship between the duty cycle D∈[0,1] and the torque (linear mapping) ; S2023, the PWM waveform V PWM (t) is expressed within the period T PWM as: ; S2024, vibration signal superposition and output, and finally the drive signal V(T') mixes two waveforms: ; Among them, α and β are weight coefficients; 2.2.4 Step S203, rigid body dynamics simulation: Use a physics engine to solve the Newton - Euler equations. After calculating the movement of the components, use spatial partitioning to screen the object pairs CPL that may collide; adopt the Separating Axis Theorem (SAT) algorithm to detect collisions, and apply the impulse method to calculate the velocities after collisions.

[0019] S2030, Spatial partitioning to screen collision pairs: S20300, BVH tree construction, recursively execute for all objects in the scene: Calculate the bounding box; Divide the object set into left and right sub - trees along the longest axis; The termination condition is that the leaf node contains ≤ k objects; S20301, Traverse the BVH tree to screen potential collision pairs: For any two nodes N 1 , N 2 , if their bounding boxes intersect, recursively detect their child nodes and screen out CPL. The screening method is as follows: ; S2031, Separating Axis Theorem collision detection: Projection calculation: For all candidate separating axes a i (including face normals and cross - product edge directions) of two objects A and B, calculate the projection intervals Proj A and Proj B : ; ; Among them, q and p are the face normal and the cross - product edge respectively.

[0020] The separation condition is: If there exists any axis a i that satisfies or , then the objects do not collide; otherwise, a collision occurs.

[0021] S2032, Impulse Method collision response: Normal impulse calculation: Let the collision normal vector be n, the restitution coefficient be e, and the relative velocity be v rel = v A - v B , the impulse ; Among them, r A , r B are the vectors from the collision point to the center of mass; v A and v B are the observed velocity and the observed - by velocity respectively. Then update the velocities after the collision: ; 2.2.5 Step S204, Trigger the state machine to roll back: Detect disassembly and assembly sequence violations (such as separating components without removing bolts) through Graph Theory, and trigger the state machine to roll back.

[0022] S2040, Graph theory modeling of the disassembly and assembly process: Let the disassembly and assembly process be a directed graph G=(V,E), where the vertex set V={v 1 ,v 2 ,...,v n} represents legal operation states (such as "power off", "bolt removed"); the edge set E⊆V i ×V j represents allowed state transitions (such as ); S2041, Current state path tracking: Define the state history stack S=[s 0 ,s 1 ,...,s t , where s k ∈V represents the state at the k-th step, and the current state is s t ; S2042, Illegal operation detection: When the user attempts to execute an action a that triggers a state transition , if the transition is illegal: (s t ,s′)∉E; then trigger the rollback condition RollbackTrigger: ; where, ∄ represents does not exist; S2043, Determine the rollback target state: Through reverse search of the history stack S, find the nearest legal predecessor state s p such that ∃(s p ,s valid )∈E, where s valid is the legal state of the current scenario.

[0023] S2044, Update the state after rollback to S←[s 0 ,s 1 ,...,s p .

[0024] 2.3 Step S3, Logic control of the disassembly and assembly process: Define disassembly and assembly steps based on a directed acyclic graph (DAG), and use breadth-first search (BFS) to push the currently executable operations (such as "the next step should be to disconnect the high-voltage interface"); at the same time, use a rule engine to monitor illegal operations in real-time and provide alerts; Use a hash table to map the association between tools and actions (such as "electric wrench → tighten bolts"), and match the user's intention through cosine similarity.

[0025] 2.3.1 Step S300, Disassembly and Assembly Step Pushing Based on DAG: Define the algorithm for this step based on the directed acyclic graph (DAG).

[0026] Let the disassembly and assembly flow chart be G=(V,E), where the vertex set V={v 1 ,v 2 ,...,v m} represents a total of m operation steps v i (such as v i = "power off", v j = "remove the housing"); the edge set E⊆V×V represents the step dependency relationship (such as E(v i ,v j )∈E means that "power off" must be done first before "removing the housing"); 2.3.2 Step S301, Generation of the Current Executable Step Set (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: ; The executable operations prompted by the user interface ; Among them, the priority order is preset based on the safety level or topological order.

[0027] 2.3.3 Step S302, Rule Engine Violation Monitoring: Let the current operation be a t , corresponding to the state transition , and the legality checking function AlarmTrigger: ; Then, when the alarm is triggered, force a rollback to the nearest legal state (see the state machine rollback algorithm above).

[0028] 2.3.4 Step S303, Tool-Action Mapping and Intention Matching The hash table stores the tool-action relationship, and defines the hash function H:K→V, where the key K is the tool type (such as "electric wrench") value, and V is the set of allowed actions, such as {tighten bolts, remove nuts}; Match user intent based on cosine similarity: Let the feature vector u=(u 1 ,u 2 ,...,u m ) (including sensor data such as torque and angle) of the current user operation, and the target action requirement vector v a =V=(v 1 ,v 2 ,...,v m ). Then the matching degree ; where ||·|| represents the logical operator "or".

[0029] Select the optimal action ; 2.4 Step S4, teaching evaluation and feedback: Record the timestamp, operation type, and component ID information, and store it in a time-series database (InfluxDB).

[0030] Based on the weighted method as the scoring rule, randomly inject faults (such as "communication interruption of the electronic control system"), and use a decision tree (ID3 / C4.5) to determine whether the trainee's troubleshooting path is correct.

[0031] 2.4.1 Step S400, operation data collection and time-series storage: When the user performs key operations, including grasping components, using tools, and completing steps, trigger data recording; the recorded fields include: (1) Timestamp: accurate to milliseconds (e.g., 2024-05-20T14:23:45.678Z); (2) Operation type: classification label (e.g., disassembling the battery case, incorrect uninsulated operation); (3) Component ID: unique identifier (e.g., a certain battery is marked as Battery.Module3.Cell25); (4) Additional data: operation parameters, tool model, and time consumption.

[0032] The information stored in the time-series database (InfluxDB) includes: (1) Data sharding: partition by trainee ID or operation stage to accelerate querying.

[0033] (2) Data compression: adopt differential encoding (DeltaEncoding) for repeated operations (mainly for continuous screwing).

[0034] (3) Label indexing: establish an inverted index for frequently queried fields (such as component ID, error type).

[0035] 2.4.2 Step S401, data flow pipeline execution: Execute real-time stream processing through Kafka or MQTT to transfer data from the VR client to the database; Duplicate removal and verification: Filter invalid operations (such as accidental triggers caused by joystick jitter).

[0036] 2.4.3 Step S402, Diagnostic logic for fault simulation: Based on predefined common faults (such as electric control communication interruption, battery cell short circuit), randomly inject faults after the user performs specific operations (such as when disassembling the electric control housing); trigger faults with probability p every fixed duration (such as 5 minutes); and simulate virtual component anomalies (such as error codes displayed on the dashboard, special effects of components smoking).

[0037] 2.4.4 Step S403, Evaluation of the trainee's troubleshooting path (decision tree logic) The order of the trainee's troubleshooting steps (such as first checking the power supply → then testing the signal line); Whether the correct detection equipment is selected (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; 2.4.4 Step S403, Decision tree training (ID3 / C4.5): Select the feature with the highest information gain (such as "whether to prioritize detecting the main control module") from the "correct / incorrect" troubleshooting paths marked in the historical data; compare the trainee's operations with the optimal path generated by the decision tree in real time.

[0038] If the trainee continuously deviates from the correct path, trigger a prompt (such as "It is recommended to check the CAN bus connection"); calculate the score based on the weighted method according to the path matching degree and troubleshooting efficiency.

[0039] (III) Mechanism for solving technical problems: (1) Enhance the interaction effect: Detect the component pointed at by the user through ray casting, calculate the interaction point coordinates, and achieve precise joystick and gesture operations. Students can walk freely in the virtual scene, observe the vehicle structure and components, and perform actual interaction operations. Detect the collision bounding box between the joystick and the component, trigger the grasping event, and determine whether the current tool is applicable through the Drools-Rete algorithm. This mechanism increases the realism and interactivity of the operation.

[0040] (2) Introduce disassembly and assembly process constraints: Record the parent-child hierarchical relationship of components through a tree-like data structure, and bind quality attributes and collision volume attributes. This provides students with a clear disassembly and assembly path and hierarchical relationship. Based on the constructed behavior tree control flow, parse the disassembly and assembly sequence and safety constraints from the configuration file. For example, when disassembling the battery, the system will set the constraint condition as "the high-voltage power must be disconnected first" to ensure that students operate in the correct sequence. Detect violations of the disassembly and assembly sequence through graph theory and trigger the state machine to roll back to the nearest legal state. This avoids disassembly failures or equipment damage caused by students' incorrect operations.

[0041] (3) Simulate torque control: Drive the handle vibration motor through a waveform generation algorithm (such as Sine / PWM) according to the torque calculated by the physics engine. This mechanism simulates the selection of torque and the tactile sensation in the actual disassembly and assembly process, cultivating students' manual operation ability and muscle reflex. Use the physics engine to solve the Newton-Euler equations and calculate the collision response after the movement of components. This further enhances the realism and credibility of the virtual scene.

[0042] In the second aspect, a VR interaction system for virtual disassembly and assembly teaching of the three-electric system of new energy vehicles, such as Figure 9 shown: (1) 3D model and scene management module: 3D model library: Stores 3D models of battery packs, motors, electric control systems and their sub-components (battery cells, BMS, inverters and cooling systems).

[0043] Scene dynamic loader: Supports the switching of different vehicle models (pure electric / hybrid) and different disassembly and assembly scenes (complete system disassembly and assembly, single-component replacement).

[0044] Hierarchical structure displayer: Supports the transparent / translucent display of the layer-by-layer decomposition of components (battery pack → module → battery cell).

[0045] (2) Physical simulation and interaction module: Rigid body dynamics engine: Simulates the mechanical interaction between tools (wrenches, lifting equipment) and components (screwing, influence of disassembly gravity).

[0046] Collision detection module: Ensures that the disassembly and assembly sequence conforms to reality (such as prohibiting the disassembly of the battery without disconnecting the high-voltage power, and unable to separate components until the bolts are completely removed).

[0047] Haptic feedback system: Simulates the vibration of tool operation through the VR handle (such as torque feedback when tightening screws).

[0048] (3) Disassembly and assembly process logic control module: Step-by-step process engine: Prompts operations step by step (such as power off → remove the outer shell → disconnect the high-voltage interface), and detects incorrect operations (such as touching high-voltage components without wearing insulating gloves) and triggers warnings / rollbacks.

[0049] Tool matching system: Automatically recommends applicable tools (such as torque wrench specifications) or prompts for incorrect tool usage based on the current step executed by the step-by-step process engine.

[0050] (4) Teaching and assessment module: Knowledge annotation module: Displays the name, function, and parameters (such as voltage, torque value) when hovering over / clicking on a component.

[0051] Fault simulation module: Implants common fault scenarios (such as battery leakage, motor overheating) for trainees to troubleshoot.

[0052] Operation scoring module: Records the operation time, number of errors, and rationality of tool usage; generates disassembly and assembly reports (such as missing steps, safety hazards).

[0053] (5) A VR device, a processor, and a memory connected to the processor, wherein program instructions are stored in the memory; The processor is connected to the above-mentioned three-dimensional model and scene management module, physical simulation and interaction module, disassembly and assembly process logic control module, teaching and assessment module, and VR device; The user interface is provided by the VR device; When the program instructions are executed by the processor, the processor is caused to execute the VR interaction method as described above.

[0054] Compared with the prior art, the beneficial effects of the present invention are: 1. Strengthen the learning and practice of disassembly and assembly processes: The present invention records the parent-child hierarchical relationship of components through a tree-like data structure and controls the disassembly and assembly process based on the constructed behavior tree to ensure that students operate in the correct order and steps. Safety rules and process configurations are built in, such as the need to disconnect high-voltage electricity before disassembling the battery, effectively avoiding safety accidents caused by incorrect operations. When a student's operation violates the regulations, the present invention triggers the state machine to roll back to the nearest legal state to help the student correct the error in a timely manner and ensure the smooth progress of the disassembly and assembly process; 2. Simulate real torque control: The present invention calculates the torque according to the physical engine and drives the handle vibration motor through a waveform generation algorithm to simulate the tactile feedback at different torques during the actual disassembly and assembly process, helping students better understand and master torque control skills. By simulating the real disassembly and assembly feel, the present invention can cultivate students' hand-operating ability and muscle reflexes, laying a solid foundation for future actual operations; 3. Improve the efficiency of teaching evaluation and feedback: The present invention can record the operation data of students in real time, including information such as timestamps, operation types, and component IDs, providing detailed data support for subsequent teaching evaluations. The present invention also supports randomly injecting faults to simulate real repair scenarios, requiring students to troubleshoot and repair faults based on the knowledge they have learned, further exercising the students' practical abilities and problem-solving abilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following described drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0056] Figure 1 It is a schematic flowchart of the method of the present invention; Figure 2 It is a schematic diagram of the sub-steps of steps S200 - S203 of the present invention; Figure 3 It is a schematic diagram of the visualization of the VR effect of the present invention; Figure 4 It is a schematic diagram of VR disassembly and torque simulation of the present invention; Figure 5 It is a schematic diagram of the physical collision effect of VR disassembly of the outer shell of the three-electric system of the present invention; Figure 6 It is a schematic diagram of the visualization effect of the plate group after VR disassembly of the outer shell of the three-electric system of the present invention; Figure 7 It is a schematic diagram of triggering the state machine rollback in VR of the present invention (prompted by a red flashing special effect); Figure 8 It is a schematic diagram of the comparison of vibration simulation effects of the present invention; Figure 9 It is a schematic diagram of the system composition of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will give a detailed description of the specific embodiments of the present invention with reference to the drawings. Many specific details are set forth in the following description to fully understand the present invention. However, the present invention can be implemented in many other ways different from those described herein. Those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below; It should be noted that the various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the embodiments, reference can be made to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description in the method section.

[0058] Embodiment 1: Traditional technologies often simply build VR demonstration scenarios. Students can only passively understand the visualized knowledge according to a fixed demonstration process, lacking actual operation and interaction. The three-electricity system of new energy vehicles is complex in structure, and specific disassembly processes need to be strictly followed during disassembly and assembly. However, traditional technologies do not add necessary constraint items to these technological steps, resulting in students being unable to learn strict disassembly and assembly techniques. During the disassembly and assembly of new energy vehicles, the removal of screws needs to strictly follow specific torques. Inappropriate torques may cause loose or damaged battery connections, affecting battery performance and lifespan. Traditional technologies can neither implement corresponding conditional constraints nor simulate the touch sensations brought about by torque selection and actual application.

[0059] Therefore, please refer to Figure 1 , this embodiment discloses a VR interaction method for virtual disassembly and assembly teaching of the three-electricity system of new energy vehicles; In this embodiment, regarding step S1, system initialization and scene loading: Load 3D models and scene configuration files, update the virtual scene in real time, construct the scene topology structure and record component relationships, and initialize safety rules and process configurations.

[0060] Specifically, in 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 floating UIs in the VR screen space; use multi-camera viewports (Multi-Viewport) or single-pass stereo rendering (Single-Pass Stereo) to support VR split-screen display, ensure correct separation of the left and right eye views, and achieve an immersive experience.

[0061] Specifically, in step S101, start the model LOD (Level of Detail) hierarchical management: Render the model only at a close distance to the camera viewport to reduce GPU load; when the model is close to the camera, use a high level of detail for rendering; when the model is far from the camera, switch to a low level of detail to reduce GPU load. By presetting different LOD models and dynamically switching according to the distance during runtime, efficient utilization of resources is achieved.

[0062] Specifically, in step S102, scene topology construction: Record the parent-child hierarchical relationship of components through a tree structure (taking the power supply system of a car as an example, its parent-child hierarchical relationship is "battery pack → module → cell"), bind quality attributes and collision volume attributes, and use quaternions and matrix transformations to manage the position and rotation of components. The tree structure makes the hierarchical relationship of components clear at a glance, facilitating students to understand the structure of the three-electric system. Attribute binding and pose management enhance the realism of interaction, enabling students to experience a near-real operation feeling in a virtual environment.

[0063] Specifically, in step S103, initialize the safety rule and process configuration: Based on the constructed behavior tree to control the process, parse the disassembly and assembly sequence and safety constraints from the configuration file (for example, when disassembling and assembling the battery, the constraint condition is "the high-voltage power must be disconnected first"). By enforcing safety constraints, it effectively prevents students' incorrect behaviors in virtual operations and cultivates their safety awareness and standardized operation habits.

[0064] It should be further pointed out that in step S1: (1) Asynchronous loading technology: Adopt Unity's Addressables or Unreal's Async Loading technology to achieve asynchronous loading of 3D models (FBX / GLTF format) and scene configuration files (JSON / XML). Allow resources to be loaded in the background and avoid blocking the main thread, thereby enhancing the user experience and ensuring that the virtual scene can start smoothly.

[0065] (2) Real-time update and prompt: During the loading process, the system updates the virtual scene in real time and prompts the user with the loading progress or relevant information through the UI, enhancing user interactivity.

[0066] (3) Shader programming and component outline light: Use Shader programming technology to achieve the component outline light effect (Outline Shader), making key components more prominent in complex scenes, facilitating students to identify and operate.

[0067] In this embodiment, as Figure 2 shown, regarding step S2, user interaction processing: Utilize the sensors built into VR devices (handles, head-mounted displays) to capture the user's hand movements and head rotations in the virtual environment in real time. Analyze the captured data to identify the user's operation intentions, including grasping, rotating, disassembling, or installing components, etc.

[0068] Through the haptic feedback devices (vibration motors, force feedback handles) of the VR device, simulate the feel during the real disassembly and assembly process, such as the resistance feeling when tightening screws, the looseness feeling when disassembling components, etc.

[0069] Apply the principle of rigid body dynamics to simulate physical phenomena such as collisions and frictions of virtual components during disassembly and assembly. When a collision occurs between components due to the user's operation, calculate the post-collision motion state and mechanical changes according to the principle of rigid body dynamics, and update the virtual scene in real time.

[0070] According to the preset disassembly and assembly sequence and safety constraints, verify in real time whether the user's operation meets the requirements. During the operation process, continuously monitor the state of virtual components and the user's operation behavior, and promptly discover and correct potential safety hazards.

[0071] Use a state machine to manage different states and stages during the virtual disassembly and assembly process, such as disassembly state, installation state, inspection state. When the user performs an illegal operation, trigger the state machine rollback mechanism, restore the virtual scene to the state before the operation, and prompt the user with the correct operation method.

[0072] Specifically, for step S200, handle / gesture input parsing: The handle detects the component pointed by the user through ray casting and calculates the interaction point coordinates; for gestures, use the MediaPipe library to identify the key points of the hand, including grasping gestures and rotating gestures.

[0073] The methods of ray casting include: S2000, generation of ray r(t): The handle ray is defined by the position O (origin) and direction d (unit vector): r(t)=O + t⋅d, t≥0; where t is the parameter at which the ray intersects each plane, and t≥0 means the ray starts from the origin and extends infinitely along the direction d.

[0074] S2001, collision detection (AABB bounding box intersection test): To efficiently detect the intersection of the ray with components in the scene, use an axis-aligned bounding box (AABB) for collision detection. The AABB is a cuboid, and its boundaries are determined by the minimum corner point b min and the maximum corner point b max Decide. For the axis-aligned bounding box (AABB) of each component in the scene, set: ; S2002, for each face of the AABB (in the x, y, z three coordinate axis directions), calculate the entry parameter t enter and the exit parameter t exit of the intersection of the ray with this face: ; where x, y, z represent the coordinate axes, and min represents its minimization parameter; In S2003, to determine whether a ray intersects with an AABB, the intersection condition needs to be met: ; That is, before the ray enters all the faces of the AABB, it must have exited a certain face of the AABB to ensure that the ray intersects with the AABB.

[0075] In S2004, when the intersection condition is met, take the nearest legal intersection point parameter t hit , and calculate the coordinates of the interaction point: ; then the coordinates of the interaction point ; It can be understood that the combination of ray casting and the intersection test of the AABB bounding box enables the handle to accurately detect the component pointed by the user, improving the accuracy and reliability of the interaction. The use of the AABB bounding box greatly reduces the computational amount of collision detection and improves the real-time performance and response speed of the interaction. Through the ray casting of the handle, the user can intuitively feel their operations in the virtual scene, enhancing the immersion and experience.

[0076] It should be noted that in step S200, the MediaPipe library is an open-source cross-platform machine learning library that provides efficient and real-time hand keypoint detection functions. By identifying hand keypoints through the MediaPipe library, the grasping gesture and rotation gesture are further parsed. These gestures can correspond to specific operations in the virtual scene, such as grasping components and rotating components. Gesture input enables the user to operate components in the virtual scene through natural gestures like in real life, improving the naturalness and intuitiveness of the interaction. Through the gesture recognition function of the MediaPipe library, multiple gestures can be input, enriching the interaction methods and means.

[0077] Specifically, in step S201, the interaction between the tool and the component: detect the collision bounding box (AABB) of the handle and the component, and trigger the grasping event; use the Drools-Rete algorithm to determine whether the current tool is applicable (such as "whether the torque wrench matches the bolt specification").

[0078] In S2010, collision bounding box detection (AABB): Based on the handle grasping point coordinates P hit =(x, y, z) (from the ray casting result), the minimum / maximum corner points of the component AABB bounding box are: ; By comparing the coordinates of the handle grasping point with the minimum and maximum corner point coordinates of the component AABB bounding box, determine whether the handle has entered the bounding box of the component.

[0079] In S2011, the determined collision condition is: ; Ensure that the handle interacts with the components at the correct position to avoid misoperation.

[0080] S2012, Determine the triggering of the grasping event: ; S2013, Use the Drools-Rete algorithm for rule matching to determine whether the current tool is applicable to the target bolt: The tool attribute TT of the currently selected tool (3D model) = (type, min torque , max torque ); where type is the encoding of the tool model, and min torque and max torque are the encoding of the minimum and maximum torque parameters of the tool respectively; In the current working environment, the target bolt specification BB = (diameter, required torque ); where diameter is the encoding of the bolt diameter parameter, and required torque is the encoding of the torque parameter required for the bolt; Rule matching (evaluation of Rete network conditions), is a logical expression: ; At the same time, more judgment rules can also be added based on the above logic of applicability judgment; Exemplarily, taking the insulation of the tool as an example: Match 安全 (TT,BB)=Match(T,B)∧(TT.Insulation level ≥ BB.Voltage level); Among them, after completing this step, the grasping of the object can be realized (for example, the tool model, as Figure 3 shown).

[0081] S2014, Rete algorithm execution: Traverse all nodes, match according to the tool attributes and bolt specifications, and find the applicable rules. Finally, activate the rule: ; where n is the number of nodes, and i is the node index; applicable is the activated rule; It can be understood that through the collision bounding box detection and grasping event triggering mechanism, it is ensured that the handle interacts with the components at the correct position to avoid misoperation. The tool applicability judgment mechanism ensures that the tool selected by the user matches the actual operation requirements, reducing the operation risk. By using the Rete algorithm to quickly find the applicable tool, it reduces the time for the user to search for the tool and improves the learning efficiency. Simulate the real disassembly and assembly process, and require the user to adjust the operation force and direction according to the feedback, so as to cultivate the hand operation ability and muscle reflex.

[0082] It should be further pointed out that the principle of the above effects is as follows: By accurately calculating the positional relationship between the handle grasping point and the AABB bounding box of the component, the accuracy of the interaction is ensured. Combining with the state of the handle trigger key, clear feedback is provided to the user to enhance the immersion of the interaction. The Drools-Rete algorithm is used for rule matching to ensure the matching of the tool and the bolt specification, improving the safety of the operation. Through an efficient node matching mechanism, the applicable tool can be quickly found, enhancing the interaction efficiency and learning effect.

[0083] Specifically, as Figure 4 shown, in step S202, haptic feedback generation: Calculate the torque of screwing the screw according to the physics engine, and drive the handle vibration motor through a waveform generation algorithm (Sine / PWM).

[0084] Therefore, the implementation of this step involves a haptic feedback generation algorithm, and the algorithm flow is as follows: S2020, mapping of torque to vibration intensity: Let the current torque calculated by the physics engine be T (unit: Nm), and define the proportional relationship between the vibration amplitude A and the torque T: ; where k is the gain coefficient, and T max is the maximum torque allowed by the system, which needs to be calibrated according to the performance of the handle motor; 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.

[0085] Among them, 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 max allowed by the system needs to be calibrated according to the performance of the handle motor to prevent motor overload. By mapping the torque to the vibration intensity, the user can perceive the change of the torque through touch, enhancing the immersion and operation accuracy.

[0086] S2021, generate a sine wave vibration signal V sine (t) related to the torque, and generate a vibration signal related to the torque using the sine wave function: ; where f is the reference frequency, set to the haptic sensitive frequency band (such as 50–200 Hz); ϕ is the initial phase; a non-linear mapping term A∝T 2 can be introduced to enhance the high-torque feedback.

[0087] S2022, define the relationship (linear mapping) between the duty cycle D∈[0,1] and the torque ; S2023, PWM waveform V PWM(t) is expressed within the period T PWM as: ; It should be noted that the duty cycle D is linearly mapped to the torque, which determines the switching time ratio of the PWM waveform.

[0088] S2024, the vibration signal is superimposed and output, and finally the drive signal V(T') mixes two waveforms: ; where α and β are weighting coefficients; It can be understood that the above PWM waveform combines the characteristics of continuous vibration and intermittent vibration by changing the ratio of the switching time and superimposing two waveforms, making the haptic feedback more rich and real. It simulates the intermittent vibration feeling of the actual tool under different torques and enhances the authenticity of the haptic feedback.

[0089] Specifically, in practice, by simulating and emulating the parameters (thread specifications, torque, equipment type) at the positions of the disassembly and assembly parts of the actual three-electric system, a comparative analysis is carried out based on the vibration sensors installed on the electric screwdriver and the dynamic deflections simulated in the VR scene. As Figure 8 shown, the solution of this embodiment can more accurately simulate the actual vibration effect (the error is about 1 - 4 mm). This indicates that the haptic feedback generation algorithm can effectively simulate the torque constraint and vibration feedback in the actual disassembly and assembly process, thereby cultivating the students' hand-operating ability and muscle reflex.

[0090] It can be understood that by simulating the torque change and vibration feedback in the actual disassembly and assembly process, users can obtain a more real operation experience in the VR environment. The haptic feedback enables users to more accurately perceive the torque change and the operation position, reducing the possibility of misoperation. Through repeated practice and the guidance of haptic feedback, users can gradually cultivate correct hand-operating habits and muscle reflexes, improving the disassembly and assembly efficiency and quality.

[0091] Specifically, for step S203, as Figure 5 shown, rigid body dynamics simulation: Use the physics engine (Phys) to solve the Newton-Euler equations. After calculating the movement of the parts, use spatial partitioning (BVH) to screen the object pairs CPL that may collide; mainly used for simulating the disassembly of objects in the scene; it uses the Separating Axis Theorem (SAT) algorithm to detect collisions and applies the Impulse Method to calculate the velocity after the collision. This series of steps is mainly used for simulating the object interaction and collision response in the disassembly scenario of the three-electric system of new energy vehicles.

[0092] S2030, spatial partitioning (BVH) screens the collision pairs: Newton-Euler equations are the fundamental equations of rigid body dynamics, used to describe the motion state of rigid bodies. These equations can be solved through existing physical engines to calculate the motion trajectories of components under external forces and provide functions such as collision detection and constraint solving to simulate the real physical world. Therefore, the core point lies in space partitioning, including the following steps: S20300, BVH tree construction, recursively execute for all objects in the scene: Calculate an enclosing box for each object in the scene to quickly judge the possible collisions between objects; Divide the object set into left and right subtrees along the longest axis until the termination condition is met (the number of objects contained in a leaf node does not exceed k); It can be understood that by constructing a BVH tree, the computational amount of collision detection can be significantly reduced. Because only pairs of objects with intersecting enclosing boxes need to be further detected, thus improving the efficiency of collision detection.

[0093] S20301, traverse the BVH tree to screen potential collision pairs. For any two nodes N 1 , N 2 , if their enclosing boxes intersect, recursively detect their child nodes and screen out the CPL. The screening method is as follows: ; Among them, for any two nodes N 1 and N 2 , if their enclosing boxes intersect, recursively detect their child nodes and screen out the possible pairs of colliding objects (CPL). The screening method makes recursive or direct judgments according to the type of the node (leaf node or non-leaf node); S2031, Separating Axis Theorem (SAT) collision detection: Projection calculation: For all candidate separating axes a i (including face normals and cross-product edge directions) of two objects A and B, calculate the projection intervals Proj A and Proj B : ; ; Among them, q and p are the face normal and the cross-product edge respectively.

[0094] The separation condition is: If there exists any axis a i satisfying or is , then the objects do not collide, otherwise a collision occurs.

[0095] It can be understood that this algorithm can accurately judge the collision state between objects, avoiding misjudgment and missed judgment.

[0096] S2032, Impulse Method collision response: In the calculation of the normal impulse, let the collision normal vector be n, the restitution coefficient be e, and the relative velocity be v rel = v A -v B , the impulse J is: ; where r A 、r B are the vectors from the collision point to the center of mass; v A and v B are the observed velocity and the observed velocity respectively. Then update the velocity after the collision: ; 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, cultivating the student's manual operation ability and muscle reflex, and improving the realism of the operation.

[0097] Then, the disassembly of the object can be achieved, as Figure 6 shown. Through rigid body dynamics simulation and collision detection response, students can observe the physical interaction effects during the disassembly process in real time. This enhances the students' sense of participation and interactivity. Students can deepen their understanding of the disassembly process of the three-electric system of new energy vehicles through practice.

[0098] Specifically, as Figure 7 shown, in step S204, trigger the state machine rollback: detect violations of the disassembly and assembly sequence through Graph Theory, and trigger the state machine rollback. First, establish a graph theory model of the disassembly and assembly process, then track the current state path, detect illegal operations, determine the rollback target state, and update the state stack, while providing visual feedback to prompt the user.

[0099] S2040, Graph theory modeling of the disassembly and assembly process: Let the disassembly and assembly process be a directed graph G=(V,E), where the vertex set V={v 1 ,v 2 ,...,v n} represents legal operation states (such as "power off", "bolt removed"); the edge set E⊆V i ×V j represents allowed state transitions (such as ); That is, each vertex represents a specific operation state, which is a legal node in the disassembly and assembly process. The edge represents a legal transition from state v i to state v j .

[0100] S2041, Tracking the current state path: Define the state history stack S=[s0 , s 1 ,..., s t , where s k ∈V represents the state at the k-th step, and the current state is s t .

[0101] S2042. When the user attempts to execute an action a, the system checks whether this action will result in a legal state transition. The illegal operation detection mechanism is as follows: Executing action a triggers a state transition , if this transition is illegal: (s t , s′) ∉ E; then the rollback condition RollbackTrigger is triggered: ; where ∄ represents does not exist; S2043. Determining the rollback target state: By reverse-searching the historical stack S, find the nearest legal predecessor state s p , such that ∃(s p , s valid ) ∈ E, where s valid is the legal state of the current scenario.

[0102] S2044. Update the state after rollback to S ← [s 0 , s 1 ,..., s p ; Preferably, simulate a red flashing special effect and bind it to the rollback target state, as Figure 7 shown, to prompt the user of the current interference point.

[0103] It can be understood that through graph theory modeling and path tracking, the system can accurately judge the legality of the user's operations, ensuring that the disassembly and assembly sequence meets the actual requirements. Avoid the disassembly and assembly failures or damages caused by the user's incorrect operation sequence, and improve the accuracy and reliability of teaching. When the user attempts to execute an illegal operation, the system can immediately detect and trigger a rollback, restoring the state to the nearest legal predecessor state. Provide immediate feedback to help the user quickly correct the incorrect operation and improve the learning efficiency and operation accuracy.

[0104] Prompt the user of the current interference point through the red flashing special effect, enhancing the user's visual perception and operation guidance. Enable the user to more clearly understand the specific location of the incorrect operation, improve the intuitiveness and interactivity of teaching, and cultivate the student's manual operation ability and muscle reflex. Moreover, by reverse-searching the historical stack to determine the rollback target state, the system can flexibly handle various illegal operation situations.

[0105] In this embodiment, regarding step S3, the disassembly and assembly process logic control: the disassembly and assembly steps are defined based on a directed acyclic graph (DAG), and the current executable operation (such as "the high-voltage interface should be disconnected in the next step") is pushed using a breadth-first search (BFS); at the same time, the rule engine is used to monitor illegal operations in real time and provide alarms; Use a hash table to map the relationship between tools and actions (such as "electric wrench → tighten bolts"), and match user intent through cosine similarity.

[0106] It should be pointed out that in this step, the DAG structure is a graph structure in which nodes represent disassembly and assembly steps, directed edges represent dependencies between steps, and there are no loops. The disassembly and assembly process of the three-electric system of new energy vehicles is decomposed into a series of steps, each step is a node of DAG, and the sequence and dependency between steps are represented by directed edges. DAG ensures the sequence and dependency of disassembly and assembly steps, avoiding the occurrence of circular dependencies and deadlocks. In DAG, starting from the current step node, the BFS algorithm is used to traverse adjacent nodes to find all currently executable operations (that is, there are no unfinished steps with predecessor dependencies). BFS ensures that executable operations are pushed in the hierarchical order of the steps, guiding users to disassemble and assemble in the correct order.

[0107] A hash table is a data structure used to quickly find and map key-value pairs. Tools (such as electric wrenches) are used as keys, and possible actions (such as tightening bolts) are stored in a hash table as values. The hash table provides fast tool-action mapping, supports real-time response to user operations, and improves the smoothness of interaction.

[0108] Cosine similarity is a measure of the similarity between two vectors, ranging from [-1, 1], with 1 indicating complete similarity. User operations (such as gestures and voice commands) are converted into vector representations and cosine similarity is calculated with predefined operation vectors. Cosine similarity matches user intent, allowing the system to understand and respond to ambiguous or incomplete user operations, improving the naturalness and flexibility of interaction.

[0109] Specifically, step S300, DAG-based disassembly and assembly step push: the algorithm of this step is defined based on a directed acyclic graph (DAG).

[0110] Assume that the assembly and disassembly flowchart G=(V,E), where the vertex set V={v 1 ,v 2 ,...,v m} indicates a total of m operation steps v i (such as v i = "Power off", v j = "Disassemble the shell"); the edge set E⊆V×V represents the step dependency (such as E(v i, v j ) ∈ E means that "power off" must be done first before "removing the outer shell"); The characteristics of a directed acyclic graph (DAG) are reflected in: (1) Directedness: The edge E(v i , v j ) represents the direction from vertex vi to vertex vj, that is, operation v i is a prerequisite for operation v j .

[0111] (2) Acyclicity: There are no cycles in the graph, that is, there does not exist a series of vertices v 1 , v 2 , ..., v k such that E(v 1 , v 2 ), E(v 2 , v 3 ), ..., E(v k , v 1 ) hold simultaneously. This ensures the sequentiality of steps and avoids deadlocks or circular dependencies.

[0112] The process of the disassembly and assembly step push algorithm based on DAG is as follows: 1) Initialization: Determine the set of currently executable operations, usually starting from vertices with no pre - dependencies (i.e., vertices with in - degree 0).

[0113] 2) Step push: 2.1) Traverse the set of currently executable operations and present the operations to the user.

[0114] 2.2) The user selects an operation and executes it.

[0115] 2.3) Update the graph G, mark the vertices of the executed operations, and remove the corresponding edges.

[0116] 2.4) Recalculate the set of currently executable operations, that is, find all unexecuted vertices with in - degree 0.

[0117] 3) Repeat: Repeat the above steps until all operations are executed.

[0118] Use an adjacency list or adjacency matrix to represent the graph G to facilitate quickly finding the adjacent vertices of a vertex and updating edge information. Maintain an in - degree array to record the in - degree value of each vertex to facilitate quickly finding vertices with in - degree 0. Each time the user executes an operation, dynamically update the graph G and the in - degree array to ensure the efficiency of the algorithm.

[0119] It is understandable that by defining the disassembly and assembly steps through a DAG, the sequentiality and dependency of the steps are ensured, avoiding operational chaos and errors. The directed and acyclic nature of the DAG guarantees the logical order between steps, preventing circular dependencies and deadlocks. The algorithm ensures that the user can only perform the currently executable operations by dynamically updating graph G and the in-degree array. The disassembly and assembly step push algorithm based on the DAG can automatically push the currently executable operations, improving the teaching efficiency and accuracy. The algorithm automatically guides the user to perform disassembly and assembly operations in the correct order through steps such as initialization, step pushing, and repeated execution, reducing human errors and omissions. By simulating real disassembly and assembly steps and dependencies, the manual operation ability and muscle reflexes of students are cultivated. The sequentiality and accuracy of the disassembly and assembly steps are ensured, enabling students to obtain an experience close to real operations in a virtual environment. Students need to perform operations in the correct order, which helps to cultivate their manual operation ability and muscle reflexes.

[0120] Specifically, in step S301, generation of the currently executable step set (BFS traversal): Define the adjacency matrix A, where A ij = 1 if and only if E(v i , v j ) ∈ E. Starting from the currently completed step set S done , find all directly reachable steps through BFS: ; The executable operations prompted by the user interface ; Among them, the priority order is preset and determined based on the safety level or topological order.

[0121] The adjacency matrix A is a two-dimensional matrix used to represent the dependency relationship between disassembly and assembly steps. The element A ij in the matrix represents 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 sequence relationship between disassembly and assembly steps can be intuitively represented, providing a basis for subsequent BFS traversal.

[0122] 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, and then sequentially visits all unvisited adjacent nodes of these adjacent nodes until all reachable nodes in the graph are visited. Starting from the currently completed step set Sdone Start from this, traverse the adjacency matrix A through BFS, and find all steps that are directly connected to S done in the previous step and have not been completed, that is, the directly reachable step set S next .

[0123] Specifically, for each step vi in Sd one , check all its adjacent steps v j (A ij = 1). If v j has not been visited (that is, not in S done ), and vj has no other preceding steps (that is, there does not exist (v k , v j ) ∈ E and v k is not in S done ), then add v j to S next . Generate the directly reachable step set S next through BFS traversal, which can ensure that students can only perform operations allowed in the current stage, avoid safety problems caused by incorrect operations, and also meet the requirements of the sequence of disassembly and assembly processes.

[0124] After determining the directly reachable step set S next , it is necessary to select an optimal step from it as the next executable operation NextStep. This step uses the priority order (Priority) as the selection criterion. The priority order is preset based on the safety level or topological order. For each step v next in S j , calculate its priority order Priority(v j ). Then, select the step with the highest priority order as NextStep. This can be achieved through a simple comparison operation, that is, find the v j that makes Priority(v j ) the largest. By introducing the priority order 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 a high safety level may involve high-voltage operations and need to be completed first; the topological order ensures that the dependency relationships between steps are correctly maintained. This mechanism helps to cultivate students' safety awareness and operation norms.

[0125] Specifically, for step S302, rule engine violation monitoring: During the virtual disassembly and assembly teaching process, each operation will cause a change in the system state. Let the current operation be at, which corresponds to the transition from the previous state s t-1 to the current state s tState transition. The system needs to track and record each operation and the resulting state transition in real time for subsequent legality verification: Assume the current operation is a t , corresponding to the state transition , and the legality verification function AlarmTrigger: ; Then, when the alarm is triggered, roll back to the nearest legal state (see the state machine rollback algorithm above).

[0126] In other words, the logic of this step is as follows: Condition 1: , 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 not, it indicates that this state transition is illegal.

[0127] Condition 2: Tool matching fails, that is, check whether the tool used in the current operation conforms to the preset tool matching rules. If not, it indicates improper tool use and is also illegal.

[0128] When the AlarmTrigger function returns 1, it indicates that the current operation is illegal; when it returns 0, it indicates that the current operation is legal. The system will trigger an alarm to remind the student that there is an error in the current operation. At the same time, to correct the error and restore to a legal state, the system will perform a state rollback operation. The state rollback operation depends on the state machine rollback algorithm mentioned above. This algorithm will roll back the system state to the nearest legal state s_{t - n} according to the historical state records, where n is the number of steps of the rollback. During the rollback process, the system will revoke all illegal state transitions and tool use records to ensure that the system restores to a legal and consistent state.

[0129] Specifically, in step S303, the tool - action mapping and intention matching hash table stores the tool - action relationship. Define the hash function H: K → V, where the key K is the tool type (such as "electric wrench") value, and V is the set of allowed actions, such as {tighten bolts, remove nuts}; Through the hash function, the set of allowed actions corresponding to a given tool type can be quickly found. The hash table allows searching with a time complexity of O(1), so as to quickly determine the set of allowed actions for a given tool type. New tool types and corresponding action sets can be added conveniently without modifying the existing logic.

[0130] Match the user's intention based on cosine similarity: Assume the feature vector of the user's current operation is u=(u 1 , u 2 ,..., u m(including torque and angle sensor data), target action requirement vector v a =V=(v 1 ,v 2 ,...,v m ), then the matching degree ; where ||·|| represents the logical operator "or".

[0131] Select the optimal action ; Through cosine similarity calculation, the intention of the user's current operation can be accurately identified, and the most compliant action can be selected. Automatically matching the user's intention and suggesting the optimal action can reduce the user's operation steps and improve the interaction efficiency. In the process of repeatedly matching and selecting the optimal action, students can gradually master the correct disassembly and assembly skills and force control methods, thereby cultivating their manual operation ability and muscle reflex. This mechanism helps students perform disassembly and assembly operations more accurately in future actual work and quickly adapt to different tools and environments.

[0132] In this embodiment, regarding step S4, teaching evaluation and feedback: Record the timestamp, operation type, and component ID information, and store it in a time series database (InfluxDB).

[0133] Based on the weighted method as the scoring rule, randomly inject faults (such as "electrical control system communication interruption"), and use a decision tree (ID3 / C4.5) to determine whether the trainee's troubleshooting path is correct.

[0134] Specifically, step S400, operation data collection and time series storage: When the user performs key operations, including grasping components, using tools, and completing steps, trigger data recording; the recorded fields include: (1) Timestamp: accurate to milliseconds (such as 2024-05-20T14:23:45.678Z); (2) Operation type: classification label (such as disassembling the battery case, incorrect uninsulated operation); (3) Component ID: unique identifier (such as a certain battery marked as Battery.Module3.Cell25); (4) Additional data: operation parameters, tool model, and elapsed time.

[0135] The information stored in the time series database (InfluxDB) includes: (1) Data sharding: partition by trainee ID or operation stage to accelerate querying.

[0136] (2) Data compression: Use differential encoding (DeltaEncoding) for repeated operations (mainly for continuous screwing).

[0137] (3) Label Index: Create an inverted index for frequently queried fields (such as component ID, error type).

[0138] Specifically, in step S401, the data flow pipeline executes: Perform real-time stream processing through Kafka or MQTT to transfer data from the VR client to the database; Filter out invalid operations (such as accidental triggers caused by handle jitter).

[0139] The method is as follows: The VR client sends the collected operation data to the message queue system in real time. The message queue system performs deduplication on the data and filters out duplicate 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.

[0140] Specifically, in step S402, the diagnostic logic of fault simulation: Based on predefined common faults (such as electric control communication interruption, battery cell short circuit), randomly inject faults after the user performs a specific operation (such as when disassembling the electric control housing); Trigger faults with probability p at fixed time intervals (such as every 5 minutes); And simulate virtual component anomalies (such as error codes displayed on the dashboard, special effects of component smoking).

[0141] Specifically, in step S403, the evaluation of the trainee's troubleshooting path (decision tree logic) The order of the trainee's troubleshooting steps (such as first checking the power supply → then testing the signal line); Whether the correct detection equipment is selected (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. The method is as follows: Extract the trainee's troubleshooting path and the corresponding "correct / incorrect" labels from historical data. Use algorithms such as ID3 / C4.5 to train a decision tree model and select the feature with the highest information gain as the judgment basis. Obtain information such as the trainee's troubleshooting steps and order, and the detection equipment selected in real time. Compare the trainee's operations with the optimal path generated by the decision tree to evaluate the trainee's troubleshooting efficiency and accuracy. If the trainee continuously deviates from the correct path, trigger a prompt (such as "It is recommended to check the CAN bus connection"). Calculate the score based on the weighted method according to the path matching degree and troubleshooting efficiency as the trainee's final evaluation result.

[0142] Specifically, in step S403, decision tree training (ID3 / C4.5): Select the feature with the highest information gain (such as "whether to give priority to detecting the main control module") from the "correct / incorrect" troubleshooting paths marked in historical data; Compare the trainee's operations with the optimal path generated by the decision tree in real time. If the trainee continuously deviates from the correct path, trigger a prompt (such as "It is recommended to check the CAN bus connection"); Calculate the score based on the weighted method according to the path matching degree and troubleshooting efficiency.

[0143] Embodiment 2: On the basis of Embodiment 1, a preferred solution for step S201 is further provided: In step S2021, by combining the interactive point coordinates P hit (previous calculation result), adjust the vibration spatial attenuation (the vibration becomes weaker as the distance increases). The solution is as follows: S20210, let the interactive point coordinates P hit =(x hit ,y hit ,z hit ), and the current position coordinates of the handle 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: ; S20211, in order to map the distance d to the vibration intensity A, an inverse attenuation function is used, where the vibration intensity A is proportional to the reciprocal of the distance d, and a base vibration intensity A 0 and an attenuation coefficient k decay are considered: ; where δ is a positive number used to control the attenuation speed. When δ is larger, the vibration intensity decays faster as the distance increases. To avoid A approaching infinity when d approaches 0.

[0144] It can be understood that in actual operation, when a tool (such as a screwdriver) is far from the part to be operated, due to the increase in the force arm and the decrease in the transmission efficiency, the vibration felt by the operator usually weakens. 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 immersion and the realism of the operation.

[0145] Vibration feedback is an important basis for the operator to judge the contact state between the tool and the part and the operation force. When the tool approaches or contacts the part, the enhancement of vibration feedback can help the operator more accurately locate the operation position and control the operation force. On the contrary, when the tool is far from the part, the weakening of vibration feedback can avoid misleading the operator and improve the operation accuracy. 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 operation state between the tool and the part according to the change of vibration feedback, so as to be more proficient in actual operation.

[0146] Embodiment 3: On the basis of step S300 of Embodiment 1, a series of dependency relationship examples for reference are further provided: Let V be the set of disassembly and assembly steps, where each element v i represents a specific disassembly and assembly step. For example: v 1 : Disconnect the high-voltage battery connection; v 2 : Remove the battery pack housing; v 3 : Take out the battery module; v 4 : Disconnect the motor power cable; v 5 : Remove the motor fixing bolts; v 6 : Take out the motor; v 7 : Disconnect the electric control system connection; v 8 : Remove the electric control module housing; v 9 : Take out the electric control module; The edge set E describes the dependencies between these steps. For example: E(v 1 ,v 2 ): It means that before removing the battery pack housing (v 2 ), the high-voltage battery connection (v1) must be disconnected first. E(v 1 ,v 3 ): It means that before taking out the battery module (v 3 ), the high-voltage battery connection (v 1 ) must also be disconnected first. E(v 4 ,v 5 ): It means that before removing the motor fixing bolts (v 5 ), the motor power cable (v4) needs to be disconnected first. E(v 5 ,v 6 ): It means that before taking out the motor (v 6 ), the motor fixing bolts (v 5 ) must be removed first. E(v 7 ,v 8 ): It means that before removing the electric control module housing (v 8 ), the electric control system connection (v 7 ) needs to be disconnected first. E(v 8 ,v 9 ): It means that before taking out the electric control module (v 9 ), the electric control module housing (v 8 ) must be removed first.

[0147] These dependencies can be drawn as a directed graph, where each node represents a disassembly and assembly step, and each directed edge represents a dependency. For example: v1→v2, v1→v3, v4→v5, v5→v6, v7→v8, v8→v9; According to this dependency graph, the planned disassembly and assembly sequence is as follows: Disconnect the high-voltage battery connection (v 1 ), remove the battery pack housing (v2 ), remove the battery module (v 3 ), disconnect the motor power cord (v 4 ), remove the motor fixing bolts (v 5 ), remove the motor (v 6 ), disconnect the connection of the electric control system (v 7 ), remove the housing of the electric control module (v 8 ), take out the electric control module (v 9 ); All of the above embodiments only express the implementation manners of the relevant actual applications of the present invention. The descriptions thereof are relatively specific and detailed, but should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several deformations and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the invention patent shall be subject to the appended claims.

[0148] 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 herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0149] Meanwhile, those skilled in the art can understand that all or part of the processes in the methods of implementing the above all embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above various methods. Among them, any reference to the memory, storage, database or other media provided in this application and used in the embodiments can include non-volatile and / or volatile memories. 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. By way of illustration and not limitation, RAM is available in many 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), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

Claims

1. A VR interactive method for virtual disassembly and assembly teaching of the three-electric system of new energy vehicles, characterized in that: The steps include: S1, loads 3D models and scene configuration files, updates virtual scenes in real time, builds scene topology and records component relationships; S2, captures and analyzes the user's operation intention in the VR environment, generates tactile feedback to simulate the real disassembly and assembly feel, simulates collision response through rigid body dynamics, and triggers state machine rollback to correct illegal operations; S3, based on the definition of disassembly and assembly steps of directed acyclic graph, uses breadth-first search to push the currently executable operations; monitors illegal operations in real time through the rule engine; uses hash tables to map the relationship between tools and actions, and matches user intent through cosine similarity; S4, records the timestamp, operation type and component ID information, and stores them as a time series database; based on the weighted method as the scoring rule, randomly injects faults, and uses the decision tree (ID3 / C4.5) to determine whether the trainee's troubleshooting path is correct.

2. The VR interaction method according to claim 1, characterized in that: The execution steps of S1 include: S100, loads 3D models and scene configuration files; updates virtual scenes and prompts information in real time, generates floating UI in VR screen space; supports VR split-screen display using multi-camera viewports; S101, start model LOD hierarchical management, so that the model is rendered only in front of the camera viewport; S102, recording the parent-child hierarchical relationship of the components through a tree data structure, binding mass attributes and collision volume attributes, and managing the component positions and postures using quaternions and matrix transformations; S103, based on building a behavior tree control process, parsing the disassembly and assembly sequence and safety constraints from the configuration file.

3. The VR interaction method according to claim 1, characterized in that: The execution method of S2 includes: S200, the handle detects the part pointed by the user through ray casting and calculates the coordinates of the interaction point, including grabbing gestures and rotation gestures; S201, detecting the collision bounding box between the handle and the component, triggering a grab event; judging whether the current tool is applicable through the Drools-Rete algorithm; S202, calculating the torque of screwing the screw according to the physical engine, and driving the vibration motor of the handle through a waveform generation algorithm; S203, using the physics engine to solve the Newton-Euler equations, after calculating the movement of the components, uses space partitioning to screen the CPL of objects that may collide; uses the separating axis theorem algorithm to detect collisions, and applies the impulse method to calculate the post-collision velocity; S204, detecting the violation of the assembly and disassembly sequence through graph theory, triggering the state machine to roll back.

4. The VR interaction method according to claim 3, characterized in that: The execution method of S200 includes: The handle ray is defined by position O and direction d: r(t)=O+t⋅d, t≥0; where t is the parameter of the ray intersecting with each plane; S2001, for the axis-aligned bounding box of each component in the scene, let: ； S2002, calculate the parameter t of the intersection of the ray and each plane enter and t exit ; Among them, x, y, z represent the coordinate axes, and min represents its minimization parameter; S2003, calculate the intersection condition: ; S2004, take the nearest legal intersection parameter t hit , calculate the coordinates of the interaction point: ； The coordinates of the interaction point ; The execution method of S201 includes: S2010, based on the handle grab point coordinates P hit =(x,y,z), calculate the minimum / maximum corner point b of the component AABB bounding box min , b max ; S2011, determine the collision condition as ; S2012, determining to trigger a grab event GrabEvent; S2013, current tool attributes TT = (type, min torque , max torque ); type is the code of the tool model, min torque and max torque The minimum and maximum torque parameters of the tool are coded respectively; target bolt specification BB = (diameter, required torque ), where diameter is the diameter parameter code of the bolt, required torque It is the torque parameter code required for the bolt; Rules match logical expressions: ； In S2014, the Rete algorithm matches tool types through nodes and associates bolt requirements, and finally activates the rules.

5. The VR interaction method according to claim 4, characterized in that: The execution method of S202 includes: S2020, assuming that the current torque calculated by the physical engine is T, and defining a proportional relationship between the vibration amplitude A and the torque T; S2021, generating a sine wave vibration signal V related to the torque sine (t); Where, f is the reference frequency; ϕ is the initial phase; S2022, PWM (Pulse Width Modulation) vibration signal generation: Define the relationship between duty cycle D∈[0,1] and torque ; S2023, PWM waveform V PWM (t) in period T PWM Inside: ； S2024, vibration signals are superimposed and output, and the final drive signal V(T') is a mixture of two waveforms: ； Among them, α and β are weight coefficients.

6. The VR interaction method according to claim 5, characterized in that: The execution method of S203 includes: S2030, space partitioning to filter collision pairs; BVH tree recursively executes for all objects in the scene: Calculate bounding box; Split the object set into left and right subtrees along the longest axis; 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 and N2, if their bounding boxes intersect, recursively check their child nodes and screen out CPL; S2031, for all candidate separation axes a of two objects A and B i , calculate the projection interval Proj A and Proj B ; Among them, q and p are the face normal and the cross product edge respectively; the separation condition is: if there is any axis a i satisfy or , then the objects do not collide, otherwise a collision occurs; S2032, based on the impulse method collision response, updating the post-collision velocity; The execution method of S204 includes: S2040, suppose the disassembly and assembly process is a directed graph G=(V,E), where the vertex set V={v1,v2,...,v n } represents a legal operation state; the edge set E⊆V i ×V j Indicates the allowed state transitions (such as ); S2041, define the state history stack S=[s0,s1,...,s t ], where s k ∈V represents the state of step k, and the current state is s t ; S2042: When the user performs action a, the state transition is triggered. , if the transfer is illegal: (s t ,s′)∉E, the rollback condition RollbackTrigger is triggered.

7. The VR interaction method according to claim 1, characterized in that: The S3 includes: S300, suppose the assembly and disassembly flow chart G=(V,E), where the vertex set V={v1,v2,...,v m } indicates a total of m operation steps v i ; The edge set E⊆V×V represents the step dependency; S301, define the adjacency matrix A, where A ij = 1 if and only if E(v i ,v j )∈E; from the currently completed step set S done Start by finding all directly reachable steps through BFS: ； Executable actions for UI prompts ; Among them, the priority order is preset based on the security level or topological order; S302, rule engine violation monitoring: Set the current operation to a t , corresponding to the state transition , check the validity function AlarmTrigger: ； Then, force a rollback to the most recent legal state when an alarm is triggered; S303, a hash table stores tool-action relationships and defines a hash function H:K→V, where the key K is the tool type value and V is the allowed action set.

8. The VR interaction method according to claim 1, characterized in that: In S4, based on the predefined faults, a fault is randomly injected after the user performs an operation; the fault is triggered with a probability p at fixed intervals; and simulate virtual component anomalies; The inspection includes: The order in which the students troubleshoot the steps; Whether the correct testing equipment is selected; Response time is the delay from when a fault occurs to when troubleshooting begins; Select the features with the highest information gain from the "correct / wrong" troubleshooting paths marked in historical data; compare the student's operation with the optimal path generated by the decision tree in real time; If the student deviates from the correct path continuously, a prompt will be triggered; the score will be calculated based on the weighted method according to the path matching degree and troubleshooting efficiency.

9. VR interactive system for virtual disassembly and assembly teaching of the three-electric system of new energy vehicles, characterized by: The system includes a processor and a memory connected to the processor, wherein program instructions are stored in the memory, and when the program instructions are executed by the processor, the processor executes the VR interaction method according to any one of claims 1 to 8.

10. The VR interactive system according to claim 9, characterized in that: include: The processor is connected to a three-dimensional model and scene management module, a physical simulation and interaction module, a disassembly and assembly process logic control module, a teaching and evaluation module and a VR device.

Images