Automobile lightweight intelligent decision database system and method based on artificial intelligence

Abstract

The invention relates to the technical field of automobile lightweight design and intelligent databases, and provides an automobile lightweight intelligent decision database system and method based on artificial intelligence. The invention aims to solve the technical problems of low design efficiency, low performance prediction and parameter optimization precision and difficulty in balancing multi-target requirements caused by data heterogeneous dispersion, difficulty in intercommunication and information splitting in the existing automobile lightweight design. According to the scheme, an intelligent decision database system is constructed through five core modules: a multi-source heterogeneous data acquisition module realizes data standardization processing and secure storage; the lightweight knowledge graph construction module establishes an association system and dynamically updates a standard; the intelligent algorithm module integrates a deep residual network, Bayesian optimization and a Transform architecture, and completes performance prediction, parameter optimization and scheme generation; the multi-objective optimization decision module screens an optimal scheme through multi-dimensional evaluation and constraint processing; the visual interaction module provides a visual interaction function and supports multi-mode deployment.

Description

Technical Field

[0001] This invention relates to the field of automotive lightweight design and intelligent database technology, specifically to an intelligent decision-making database system and method for automotive lightweighting based on artificial intelligence. Background Technology

[0002] In response to the demands for energy conservation, emission reduction, and performance improvement, lightweight design has become a core development direction in the automotive industry. However, the current lightweight design process faces multiple technical bottlenecks, including: First, lightweight-related data are complex and scattered, encompassing 3D models, test reports, process documents, etc., belonging to different systems. This leads to format incompatibility, difficulties in data interoperability, and insufficient security protection, making efficient data reuse difficult. Second, design information such as processes, materials, patents, and standards is fragmented, requiring manual retrieval and integration, which is inefficient and prone to overlooking key relationships, affecting design accuracy. Third, material performance prediction relies on experience or simple models, process parameter optimization uses trial and error, and scheme generation requires repeated manual adjustments, resulting in low accuracy, long cycles, and significant uncertainty. Fourth, designs often focus on a single objective, neglecting multi-dimensional requirements, and suffer from low constraint processing efficiency and difficulty in identifying the optimal solution, making it difficult to meet comprehensive design requirements.

[0003] Therefore, an AI-based intelligent decision-making database system and method for automotive lightweighting are proposed. This system achieves data standardization and secure storage through a multi-source heterogeneous data acquisition module; establishes an information association system and dynamically updates standards through a lightweight knowledge graph construction module; realizes performance prediction, parameter optimization, and automatic solution generation through an intelligent algorithm module; balances multi-dimensional requirements and selects the optimal solution through a multi-objective optimization decision-making module; and optimizes the user experience and supports flexible deployment through a visualization interaction module, providing a full-process intelligent solution for automotive lightweighting design. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent decision-making database system and method for lightweight vehicles based on artificial intelligence.

[0005] To achieve the above-mentioned objectives, the present invention provides the following technical solution: The AI-based intelligent decision-making database system for lightweight vehicles includes the following modules: The multi-source heterogeneous data acquisition module is used to standardize the processing of automotive lightweighting-related data through a reverse engineering data conversion interface, experimental data cleaning components and a dynamic data labeling system, and to manage the data through a distributed storage unit and a safety control unit. The lightweight knowledge graph construction module is used to extract entities and relationships in the field of automotive lightweighting based on the BERT text parsing model, store the constructed knowledge graph through a graph database, and configure industry standard update units to maintain the timeliness of the graph. The intelligent algorithm module integrates a material performance prediction submodule, a process parameter optimization submodule, and a scheme generation submodule. The material performance prediction submodule adopts a deep residual network architecture, the process parameter optimization submodule is based on a Bayesian optimization algorithm, and the scheme generation submodule adopts a Transformer architecture. The multi-objective optimization decision module is used to process constraints and output optimization solutions based on a multi-dimensional evaluation index system that includes basic performance, economy, environmental protection, reliability and compliance, through a feasible domain search algorithm. The visualization and interaction module provides material performance comparison, 3D preview of solutions, and data export functions, and supports local, cloud, and hybrid deployment modes.

[0006] Preferably, the reverse engineering data conversion interface of the multi-source heterogeneous data acquisition module is configured with a parameter extraction algorithm based on the OpenCASCADE framework, which includes the following steps: Read the geometric topology of the 3D digital model and obtain the coordinate information of faces, edges and vertices; Match a preset feature point template, the template containing feature parameters such as sheet metal thickness threshold of 0.5-5mm, hole diameter deviation of ±0.1mm, and hole position coordinate accuracy of ±0.05mm; Automatically extract structural dimensions, material identification, and assembly constraints of components; Output a standardized digital model in STEP format.

[0007] Preferably, the distributed storage unit adopts the Hadoop Distributed File System, which includes one master node, one standby node and at least eight data nodes. The switching condition between the master node and the standby node is that the master node has no response time for more than 30 seconds. The single-node storage capacity of the data nodes is not less than 10TB, the data block size is 128MB and the number of replicas is 3.

[0008] Preferably, the safety control unit includes: The access control component based on the RBAC model is configured with permission scopes for three roles: engineers, administrators, and suppliers. Engineers only have the right to view and edit data within their own department, administrators have the right to manage all data, and suppliers only have the right to view authorized process data. The data encryption component uses the AES-256 encryption algorithm to encrypt structured data and the SSL / TLS 1.3 protocol to encrypt unstructured data during transmission. The data masking component automatically masks data shared outside the system based on regular expressions.

[0009] Preferably, the BERT text parsing model of the lightweight knowledge graph construction module adopts the bert-base-chinese pre-trained model, which has been fine-tuned and trained with 8,000 labeled documents, including patent documents, process manuals and industry standards; the model's entity extraction F1 score is not less than 0.92, the relationship recognition accuracy is not less than 0.90, and the extracted core relationships include the relationship between process and applicable materials, and the relationship between materials and prohibited or restricted substances.

[0010] Preferably, the material performance prediction submodule adopts a ResNet-50 deep residual network, which contains 50 3×3 convolutional layers and 16 residual blocks. The input features are 28-dimensional, including 8 basic material properties, 6 operating conditions, and 14 historical performance parameters. The output dimension is 2, corresponding to the strength decay value and the number of lifetime cycles, respectively. The training dataset of the submodule contains more than 200,000 sets of test data covering 32 types of automotive materials, with the test operating temperature range from -40℃ to 120℃.

[0011] Preferably, the Bayesian optimization algorithm of the process parameter optimization submodule uses the RBF kernel function with a noise coefficient of 0.1. The training data of this submodule includes more than 20,000 sets of historical process data, which includes the parameter range of thermoforming temperature 500-900℃ and pressure 10-50MPa. The optimization objective is to maximize the yield weight of 0.6 and minimize the forming time weight of 0.4. The iterative process includes parameter initialization, objective function value calculation, Gaussian process model update and parameter selection steps, with 30 iterations.

[0012] Preferably, the multi-dimensional evaluation index system of the multi-objective optimization decision-making module includes: The basic performance index has a weight of 0.3 and includes a weight score and a strength score. The weight score is calculated by subtracting the actual value from the upper limit value and dividing by the upper limit value, then multiplying by 30. The strength score is calculated by dividing the actual value by the required value and multiplying by 30. The economic indicator has a weight of 0.25 and includes material cost score and process cost score. The material cost score is calculated by subtracting the actual value from the budget and dividing by the budget, then multiplying by 25. The process cost score is calculated by subtracting the actual cost from the benchmark cost and dividing by the benchmark cost, then multiplying by 25. The environmental performance index has a weight of 0.2. The carbon emission score is calculated by subtracting the actual value from the benchmark value and dividing by the benchmark value by 20 points. The benchmark value is the carbon emission value of the original steel scheme. The reliability index has a weight of 0.15, and the fatigue life score is calculated by dividing the actual value by the required value and multiplying by 15. The compliance indicator has a weight of 0.1, which includes a patent risk score of 5 points for low risk, 3 points for medium risk, and 0 points for high risk, as well as a compliance score of 5 points for compliant and 0 points for non-compliant substances.

[0013] Preferably, the 3D simulation preview unit of the visualization interaction module integrates the Three.js engine and supports the following functions: Import the STEP format digital model and generate a 3D mesh model; The model can be rotated 360 degrees by dragging with the mouse, zoomed from 1 to 5 times by scroll wheel, and viewed by section lines; Collision detection is performed on areas with gaps ≤0.5mm, and these areas are highlighted in red. It can label the strength and weight parameters of parts and supports dragging and adjusting the position of the parameter labels.

[0014] An AI-based intelligent decision-making method for vehicle lightweighting, applied to the aforementioned AI-based intelligent decision-making database system for vehicle lightweighting, includes the following steps: S1 Multi-Source Heterogeneous Data Standardization Acquisition and Secure Storage: A 3D model analysis tool is used to analyze the geometric topology of different versions of automotive component 3D models, extracting structural dimensions, material identifiers, and assembly constraints to generate standardized model files. A temporal neural network model is used to perform trend fitting and outlier removal on automotive material bench test data. A semi-supervised learning algorithm is used to automatically annotate unstructured documents related to automotive lightweighting and generate structured data. The standardized data is stored in a distributed data storage system, and data security management is achieved through hierarchical access control, data encryption, and transmission encryption mechanisms. S2 Lightweighting Knowledge Graph Construction: Based on a pre-trained language model, patent documents, process manuals, and industry standard texts in the field of automotive lightweighting are processed to extract four types of entities: "materials, processes, patents, and standards," as well as the relationships between these entities. A graph database is used to store these entities and relationships to construct a knowledge graph. The knowledge graph contains nodes with unique identifiers and source information, as well as relationship edges with association strength parameters. A dynamic update mechanism for the knowledge graph is established to periodically acquire the latest industry standards and update the graph content accordingly. S3 Hybrid Intelligent Algorithm-Driven Solution Generation: By inputting material properties, operating conditions, and historical performance characteristics data through a deep residual network, it predicts the performance degradation law and fatigue life of the target material under different operating conditions; based on a Bayesian optimization algorithm and constrained by preset multi-objective optimization indices, it iteratively optimizes the processing parameters of automotive parts; utilizing the Transformer architecture to input vehicle structure data and design constraints, it generates multiple lightweight design solutions including material replacement, structural adjustment, and process routes; and establishes a model iteration and update mechanism to adjust the parameters of the prediction and optimization models based on newly added sample data. S4 Multi-Objective Optimization Decision Screening: A multi-dimensional evaluation system is constructed, encompassing basic performance, economy, environmental friendliness, reliability, and compliance. Pre-defined weights are used to comprehensively score each design scheme. Schemes exceeding constraints are eliminated using a feasible region search algorithm, and the remaining schemes are validated using a particle swarm optimization algorithm. A dimensionality reduction algorithm is employed to map the multi-dimensional optimization space to a low-dimensional plane and label the Pareto optimal solution set. The impact of key variables on target indicators is analyzed based on variance decomposition to determine optimization priorities. S5 Visualization Interaction and Result Output: Generates material performance parameter comparison charts based on data visualization tools, supports parameter weight adjustment and sorting operations; realizes the 3D model display of design schemes through a 3D simulation engine, and provides model rotation, scaling and interference detection functions; supports exporting scheme parameter tables and decision analysis reports in preset formats, adaptable to local deployment, cloud deployment or hybrid deployment modes.

[0015] Compared with existing technologies, this invention provides an intelligent decision-making database system and method for lightweight vehicles based on artificial intelligence, which has the following beneficial effects: 1. In this solution, a multi-source heterogeneous data acquisition module is used to achieve standardized processing of various types of data. At the same time, it is equipped with distributed storage and multi-layer security mechanisms, which not only eliminates the barriers to data exchange between different sources, but also ensures the security of data storage and transmission. This transforms scattered data resources into standardized data assets that can be uniformly accessed, providing a reliable data foundation for the subsequent lightweight design process and breaking through the bottlenecks of low data utilization and high security risks in the traditional data processing mode.

[0016] 2. This solution utilizes a lightweight knowledge graph construction module to extract and store the core relationships between multi-dimensional information, while dynamically updating industry standards to form an integrated knowledge system encompassing "process-materials-patents-standards." During the design process, related information can be directly accessed without fragmented searches, significantly reducing information integration time, avoiding design deviations caused by information fragmentation, and improving the accuracy and efficiency of design decisions.

[0017] 3. In this solution, an intelligent algorithm module employs deep residual networks, Bayesian optimization, and Transformer architecture to achieve accurate prediction of material properties, scientific optimization of process parameters, and automatic generation of design schemes, while also supporting multi-scenario template adaptation. Compared to traditional methods, this module does not rely on human experience, reducing performance prediction bias, avoiding blind parameter optimization, shortening the scheme generation cycle, and enabling lightweight design to shift from "experience-driven" to "data and algorithm-driven," thereby improving the accuracy and efficiency of design results.

[0018] 4. In this solution, a multi-objective optimization decision-making module is used to establish a multi-dimensional evaluation system covering basic performance, economy, environmental protection, reliability, and compliance. Combined with the feasible region fast search algorithm and Pareto front visualization function, it can efficiently handle multiple constraints, accurately select the optimal solution that balances multiple objectives, avoid design defects caused by single objective optimization, and ensure that the solution meets performance requirements while taking into account cost control, environmental protection requirements, and compliance, thus adapting to the comprehensive needs of automotive lightweight design. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the system framework of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 The AI-based intelligent decision-making database system for lightweight vehicles includes the following modules: 1. Multi-source heterogeneous data acquisition module This module serves as the system's "data entry point." Addressing the challenges of complex data types (3D models, test reports, process documents, etc.) and fragmented storage (belonging to different systems of OEMs and suppliers) in automotive lightweight design, it standardizes data through three main components: a reverse engineering data conversion interface, an experimental data cleaning component, and a dynamic data annotation system. It is also equipped with distributed storage and security mechanisms.

[0022] 1.1 Core Functional Components Reverse engineering data conversion interface: Supports parsing of mainstream 3D modeling software (V5-V6, 1847-2212, etc.) formats. It incorporates a parameter extraction algorithm based on the OpenCASCADE open-source framework. The algorithm execution steps are: ① Read the geometric topology of the 3D model (including face, edge, and vertex coordinates); ② Match a preset template of 128 feature points (template parameters: sheet metal thickness threshold 0.5-5mm, hole diameter deviation ±0.1mm, hole position coordinate accuracy ±0.05mm); ③ Automatically extract component structural dimensions (such as plate thickness, hole diameter, and contour dimensions), material identifiers (such as material grade fields), and assembly constraints (such as bolt connections and welding relationships); ④ Output a standardized STEP format model. Testing shows that this interface achieves a parsing success rate of ≥98% for the above-mentioned model formats, with a parameter extraction error ≤0.02mm. It can solve data interoperability problems caused by incompatibility with traditional software. Those in the relevant technical field can reproduce the interface function based on the OpenCASCADE framework and the above steps.

[0023] Experimental data cleaning component: An anomaly identification model is built based on a Long Short-Term Memory (LSTM) network. The model structure and training process are as follows: ① The input layer dimension is set to 64 (corresponding to 64 experimental data features, such as strain value, temperature, and sampling time); ② There are 3 hidden layers, with 128, 64, and 32 neurons in each layer, and the activation function is ReLU; ③ The optimizer is Adam, the learning rate is 0.001, and the number of training iterations is 200 rounds; ④ The training dataset consists of 100,000+ sets of automotive material test data (covering 32 types of materials such as AA6061-T6 and DP980, including abnormal data samples such as sensor failure and temperature interference). When applying the model, trend fitting is performed on the strain data from bench tests (sampling frequency 100Hz) and fatigue test life data. When the data deviates from the fitted curve by ±3σ, it is judged as abnormal. It can accurately identify scenarios such as sensor failure (strain value jump > 50MPa) and environmental interference (error caused by temperature fluctuation ±10℃). The anomaly identification accuracy rate is over 96%, and the pass rate after data cleaning is 99%. The above model structure, training parameters and application logic are clear and can be repeatedly implemented by technicians.

[0024] The dynamic data annotation system employs the LabelSpreading semi-supervised learning algorithm. The specific implementation steps are as follows: ① Select 10% of the unstructured data (500 documents in total, including text test reports and PDF process documents) as "seed data," and manually label them with "material type (32 categories), process name (18 categories), and test item (24 categories, including tensile strength and flexural modulus)"; ② Initialize the algorithm model, setting the regularization parameter to 0.2 and the neighborhood size to 5; ③ Automatically classify and label the remaining 90% of the unstructured data. During the annotation process, manual correction of annotation errors is allowed every 10 iterations (the correction results are fed back to the model to optimize accuracy); ④ The final annotation results are stored in JSON format (fields include "data ID, annotation label, and confidence level"), with an annotation accuracy consistently above 94%. The algorithm's parameter settings, iteration process, and manual interaction mechanism are clearly defined and meet the feasibility requirements.

[0025] 1.2 Storage and Security Mechanisms Distributed Data Lake Architecture: Utilizing the Hadoop Distributed File System (HDFS), the specific deployment scheme is as follows: ① One NameNode primary node and one backup node (primary / backup failover trigger condition: primary node unresponsive for more than 30 seconds); ② Eight DataNodes, each with a storage capacity of 10TB, a data block size of 128MB, and a replica count of 3 (to avoid single-point data loss); ③ Data read / write process: Client requests data block location from NameNode → reads / writes data from DataNode → NameNode updates metadata. This architecture supports petabyte-level data expansion, with a data read / write rate ≥100MB / s, and can meet the storage needs of 200,000+ lightweight data samples (including 3D models and experimental data). The deployment steps and parameters are clearly defined and can be implemented by technical personnel.

[0026] Multi-layered data security mechanism (1) Access control: Based on the RBAC (Role-Based Access Control) model, the role and permission correspondence is as follows: Engineer (can view / edit data of this department, but has no deletion permission), Administrator (manages all data, including permission allocation), Supplier (can only view authorized process data, but has no annotation permission); Operation logs are stored in an independent database, and the record content includes "operator ID, operation time, data identifier, operation type", and the retention period is ≥1 year.

[0027] (2) Encrypted data storage: Structured data (material performance parameters) adopts the AES-256 encryption algorithm. The encryption steps are as follows: ① Generate a 256-bit key (randomly generated by the administrator through the system background, and the key is backed up to the offline device); ② Adopt CBC encryption mode, with an initial vector of 16 bits (randomly generated); ③ The encrypted data is stored in the MySQL database; Unstructured data (3D model) is encrypted using the SSL / TLS1.3 protocol during transmission, and local storage adopts file-level AES-256 encryption.

[0028] (3) Data desensitization processing: Configure desensitization rules based on regular expressions, such as replacing “cost ¥\d+.\d+ / kg” with “cost ¥XX / kg” and keeping the first 10 characters of “patent number CN\d+.\d+” (hiding the last 4 characters); Desensitization trigger conditions: Automatically executed when sharing data with external systems, and not desensitized when using data internally.

[0029] 2. Lightweight Knowledge Graph Construction Module This module serves as the system's "knowledge hub," using natural language processing technology to construct a "process-material-patent-standard" correlation graph, thus solving the problem of information fragmentation in traditional design.

[0030] 2.1 Knowledge Extraction and Construction Text parsing model: A BERT-based natural language processing model is adopted. Specific technical details are as follows: ① The pre-trained model uses bert-base-chinese (110M parameters, open-source and available); ② The fine-tuning dataset consists of 8000 annotated documents (including 4000 patent documents, 2000 process manuals, and 2000 industry standards), with annotations including "entity type (material, process, patent number, standard number), entity boundary, and entity relationship"; ③ Model training parameters: batch size 32, learning rate 2e-5, 50 iterations, and cross-entropy loss function; ④ Entity extraction and relationship recognition process: text segmentation → part-of-speech tagging → entity recognition (F1 score 0.92) → relationship classification (accuracy 0.90), extracting 6 core relationships such as "process-applicable materials," "material-prohibited and restricted substances," "process-related patents," and "material-compliant standards." The above model selection, training data, parameters, and process are clearly defined and highly reproducible.

[0031] The graph storage structure uses Neo4j graph database (version 5.10, supporting distributed deployment). The specific storage scheme is as follows: ① Node types are divided into four categories: "Materials," "Processes," "Patents," and "Standards." Node attributes include "Unique ID (format: type_serial number, e.g., MAT_001), name, category, creation time, and source (e.g., patent number, standard number)"; ② Relationship edge types correspond to six core relationship types, with attributes including "association strength (0.1-1.0, calculated as: literature citation frequency × 0.6 + expert score × ...)". 0.4 (expert scoring range 0-1), update time, and correlation basis (e.g., patent paragraph number: CN202310XXXXXX_3.1); ③ Example ternary: <Process_Thermoforming-Applicable Materials-Material_Aluminum-Silicon Coated Steel Plate> (correlation strength 0.95, correlation basis: Section 3.1 of patent CN202310XXXXXX), <Material_Cadmium-Containing Electroplated Bolts-Restricted Standards-Standard_RoHS2.0> (correlation strength 1.0, correlation basis: Article 4 of Appendix II of RoHS2.0). The node / relationship definitions and attribute rules of this storage structure are clear, and technicians can build the graph based on Neo4j.

[0032] 2.2 Map Update and Interaction Industry Standard Database and Update Mechanism: ① Built-in 36 domestic and international automotive industry standards, categorized into "Material Compliance (12 standards, such as RoHS 2.0), Process Specifications (15 standards, such as GB / T32465-2015), and Environmental Requirements (9 standards, such as ISO14067-2018)". The standard content is stored in a structured format (fields include "standard number, standard name, clause number, requirement description, effective date, and revision history"); ② Update triggering methods: Regular updates (automatically crawling the websites of the National Standardization Management Committee and the European Union ECE on the 1st of each month, with the crawling scope limited to "automotive materials and process-related standards"), and manual updates (administrators upload new standard documents in PDF format → system calls OCR to recognize text → parse structured information → automatically supplement triples); ③ Post-update processing: The graph relationship edges automatically synchronize the "effective date and revision version" attributes, and old version standards are marked as "historical versions" (not deleted, retaining traceability function). The triggering conditions, crawling scope, and parsing process of this update mechanism are clear, with no technical ambiguities.

[0033] Visual interactive features: ① The interactive interface is developed based on ECharts. The node display rules are: diameter 5-20px (the larger the diameter, the more related entities), and color to distinguish types (materials: blue, processes: green, patents: yellow, standards: red); ② Operation modes: keyword search (supports fuzzy matching, such as entering "zinc-nickel" to locate "materials_zinc-nickel alloy"), node association click (after clicking a node, the directly related 3-level nodes and relationship edges are automatically displayed, and the association information includes "applicable processes, related patents, compliance, and alternative materials"); ③ Information traceability: patent nodes display "patent number, application date, and expiration date", and standard nodes display "standard number, revision date, and key clauses". Clicking "patent number" will jump to the system's built-in patent abstract page (the abstract content comes from the public patent database). The interface development tools, node rules, and operation process of this interactive feature are clear and can be implemented by technical personnel.

[0034] 3. Intelligent Algorithm Module: The core computing unit, enabling intelligent processing throughout the entire "prediction-optimization-generation" process. This module, acting as the "brain" of the system, integrates a hybrid algorithm framework to provide intelligent solutions to problems such as "unknown material properties, difficulty in determining process parameters, and inefficient scheme design."

[0035] 3.1 Submodule Algorithm Material performance prediction submodule: ResNet-50 deep residual network is adopted. The specific technical solution is as follows: ① Model structure: It contains 50 convolutional layers (3×3 kernel size), 16 residual blocks, and the output layer is a fully connected layer (output dimension 2, corresponding to "strength decay value, life cycle number"); ② Input features: 28 dimensions (8 basic material properties: composition, density, hardness; 6 working conditions: temperature, vibration frequency, pressure; 14 historical performances: tensile strength, yield strength, fatigue life, etc.); ③ Training dataset: 200,000+ sets of experimental data (covering 32 types of automotive materials, including working conditions from -40℃ to 120℃, data source is laboratory bench testing), the data preprocessing steps are "normalization (Min-Max normalization to [0,1]) → missing value filling (mean filling) → data splitting (80% training set, 20% validation set)"; ④ Prediction process: input material parameters → model forward propagation → output performance decay curve (sampling point interval 100 hours, including strength and stiffness decay values). Example: Input AA6061-T6 aluminum alloy (thickness 3mm, temperature 80℃, vibration frequency 200Hz), output "strength decays to 480MPa after 100,000 cycles", prediction error ≤5%. The model's structure, input / output definitions, training data, and process are clearly defined, and it can be repeatedly trained and applied.

[0036] The process parameter optimization submodule is based on a Bayesian optimization algorithm. Specific implementation details are as follows: ① The kernel function is RBF (Radial Basis Function), with a noise figure set to 0.1; ② Training data: 20,000+ sets of historical process data (hot forming temperature 500-900℃, pressure 10-50MPa, corresponding to yield rates and forming times for 12 part types); ③ Optimization objectives: Maximize yield rate (weight 0.6) and minimize forming time (weight 0.4); ④ Iterative process: Initialize 5 sets of parameters → calculate the objective function value → update the Gaussian process model → select the next set of parameters (based on the expected improvement criterion) → output the optimal parameters after 30 iterations. Example: For hot-formed parts, the output is "temperature 750℃, pressure 35MPa, holding time 3min", corresponding to a yield rate of 98%. The kernel function selection, objective function definition, and iterative steps of this algorithm are clearly defined and can be implemented by technicians through programming.

[0037] The scheme generation submodule is based on the Transformer architecture. The specific technical solutions are as follows: ① Model structure: 6 layers encoder, 6 layers decoder, 8 attention heads, 512 hidden layer dimensions, and GELU activation function; ② Input parameters: Vehicle structure model (including dimensions and stress point coordinates, e.g., SUV body 4600×1880×1650mm), design constraints (weight ≤1500kg, strength ≥500MPa, crash rating ≥C-NCAP 5 stars); ③ Scheme generation process: Input parameter encoding → encoder feature extraction → decoder generating scheme fragments → splicing into a complete scheme (including "material replacement list, structural topology parameters, and process route"); ④ Output format: 3-5 schemes, each containing "part name, raw material, replacement material, thickness adjustment value, and process type". Example: Output "Door inner panel: steel → AA6061-T6 (thickness 2.5mm → 2.0mm), process: laser welding". The model's structure, input / output format, and generation process are clearly defined, without any ambiguous technical descriptions.

[0038] 3.2 Model Iteration and Scene Adaptation Model Iterative Optimization Mechanism: ① Iteration Cycle: Once per quarter; ② New Data: 10,000+ new samples (including new material test data, such as fatigue data of carbon fiber composites; new process parameters, such as hydraulic forming pressure values); ③ Incremental Training Steps: Freeze 50% of the model's bottom layer parameters → Load new data → Fine-tune the upper layer parameters (learning rate 0.0001, 20 iterations) → Validation set evaluation (if the accuracy improves by ≥1%, save the new model); ④ Optimization Effect: Material property prediction error decreases by 1%–2% per quarter. The cycle, data sources, and training steps of this iterative process are clearly defined and can be executed regularly.

[0039] Multi-scenario adaptability: Three types of scenario templates are preset, and the template parameters can be adjusted through the administrator interface (weight range 0-1): ① New energy vehicle battery pack: fire resistance weight 0.4, weight reduction weight 0.3, cost weight 0.2, corrosion resistance weight 0.1, and material performance prediction adds parameters "flame retardant rating (UL94V0)" and "thermal conductivity (≤0.2W / (m·K)"); ② Vehicle body structure: strength weight 0.35, cost weight 0.3, weight weight 0.2, welding deformation weight 0.15, and process optimization adds constraints "welding strength ≥350MPa, deformation ≤0.5mm"; ③ Fasteners: corrosion resistance weight 0.3, assembly reliability weight 0.25, cost weight 0.25, weight weight 0.2, and performance prediction adds the parameter "salt spray test life ≥500h". The template weight settings and new parameters are clearly defined, allowing for flexible adaptation to different scenarios.

[0040] 4. Multi-objective optimization decision-making module: Balances multi-dimensional requirements and selects the optimal solution. This module acts as the system's "decision filter," using multi-constraint processing, Pareto optimization, and sensitivity analysis to select the optimal solution.

[0041] 4.1 Assessment and Constraints Multi-dimensional evaluation indicator system (1) Basic performance indicators (weight 0.3): ① Weight: scored according to “(upper limit value - actual value) / upper limit value × 30 points” (e.g., if the upper limit is 1500kg and the actual value is 1480kg, the score is 28 points); ② Strength: scored according to “actual value / required value × 30 points” (e.g., if the required value is 500MPa and the actual value is 520MPa, the score is 31.2 points). The evaluation is based on the calculated data value and the material performance curve.

[0042] (2) Economic indicators (weight 0.25): ① Material cost: scored according to “(budget - actual value) / budget × 25 points” (e.g., budget 8500 yuan, actual 8200 yuan, 24.1 points); ② Process cost: scored according to “(benchmark cost - actual cost) / benchmark cost × 25 points”, the evaluation basis is the supplier quotation database (100+ suppliers publicly quoted) and the process cost model (including equipment depreciation and labor cost calculation formulas).

[0043] (3) Environmental indicators (weight 0.2): Carbon emissions (unit kgCO2) are scored according to “(benchmark value - actual value) / benchmark value × 20 points”. The benchmark value is the carbon emissions of the original steel scheme. The assessment basis is the material carbon footprint database under ISO14067 standard (full life cycle data of 120+ materials).

[0044] (4) Reliability index (weight 0.15): Fatigue life is scored by “actual value / required value × 15 points” (e.g., required 100,000 cycles, actual 120,000 cycles, 18 points are obtained), and the evaluation basis is the output value of the prediction model.

[0045] (5) Compliance indicators (weight 0.1): ① Patent risk: low (5 points), medium (3 points), high (0 points), based on the patent validity and infringement risk assessment in the knowledge graph; ② Restricted substances: compliant (5 points), non-compliant (0 points), based on RoHS 2.0 and other standards, non-compliance triggers the recommendation of alternative materials.

[0046] The evaluation results are based on a 100-point scale, and the clear scoring rules ensure that technicians can perform repeated calculations.

[0047] Multi-constraint processing unit: Employs a "fast feasible region search algorithm," with the following steps: ① Input nonlinear constraints (e.g., strength ≥ 500 MPa, cost ≤ 8500 yuan, carbon emissions ≤ 1200 kg CO2, fatigue life ≥ 100,000 cycles); ② Boundary detection: Calculate the boundary values ​​of each constraint, eliminating 80% of solutions that exceed the boundaries (e.g., cost exceeding 10,000 yuan), taking ≤ 10 seconds; ③ Refined verification: Employ a particle swarm optimization algorithm (50 particles, 20 iterations) to verify whether the solution satisfies all constraints, taking ≤ 30 seconds; ④ Output a set of solutions that meet the constraints, achieving a 100% constraint satisfaction rate. The algorithm's steps and parameters are clearly defined, and constraint verification can be implemented programmably.

[0048] 4.2 Optimization and Analysis Pareto Front Visualization Component: Employs the t-SNE dimensionality reduction algorithm. Specific implementation details are as follows: ① Parameter settings: perplexity=30, iterations=1000, learning rate=200; ② Dimensionality reduction process: Maps the 5-dimensional optimization space (weight, cost, carbon emissions, intensity, lifetime) to a 2-dimensional plane. The horizontal axis represents the "weight-cost composite value (weights of 0.5 each)," and the vertical axis represents the "intensity-lifetime composite value (weights of 0.5 each)"; ③ Visualization presentation: Each point represents a solution (labeled with the solution number and comprehensive score). The Pareto front is marked with a solid red line (the set of non-dominated solutions, i.e., the set where improving one objective does not necessarily decrease the other). Example: Solution A (1480kg, 8200 yuan, 89 points) and Solution B (1500kg, 7800 yuan, 87 points) are both non-dominated solutions. Clicking on a point displays detailed parameters. The algorithm's parameters, dimensionality reduction process, and visualization rules are clearly defined, and the visualization effect can be reproduced.

[0049] Sensitivity analysis tool: Based on variance decomposition, the specific analysis steps are as follows: ① Determine the analysis variables (material thickness 3-5mm, process temperature 700-800℃, structural dimensions 100-150mm) and target indicators (weight, cost); ② Calculate the variance contribution of each variable (formula: variance of a variable / total variance × 100%); ③ Output results: displayed as a bar chart, with the horizontal axis representing the variable and the vertical axis representing the percentage contribution. Example: Material thickness contributes 65% to weight, and process temperature contributes 40% to cost. This method has clear formulas and variable selection, and the analysis results can be quantified.

[0050] 5. Visual Interaction Module: User interface that intuitively presents the "data-solution" relationship. This module serves as the system's "user window," providing intuitive interactive functions and adapting to multiple deployment scenarios.

[0051] 5.1 Core Interactive Functions Material Performance Comparison Heatmap: Developed using ECharts, the specific implementation scheme is as follows: ① The horizontal axis represents 32 types of materials (AA6061-T6, DP980, etc.), and the vertical axis represents 5 parameters (strength, density, cost, carbon emissions, fatigue life); ② Color mapping rules: parameters are normalized to [0,1], red=1 (optimal), blue=0 (worst), and the color gradient is divided into 10 levels; ③ Interactive functions: hovering the mouse displays "material name + specific parameter value" (e.g., AA6061-T6: strength 520MPa, cost 25 yuan / kg); parameter weight adjustment (drag the slider to change the weight, e.g., when the cost weight is adjusted to 0.5, the color contrast of the cost column is increased by 200%); sorting by parameter (clicking the parameter name on the vertical axis sorts the materials in descending / ascending order according to that parameter). The development tools, mapping rules, and interactive functions of this heatmap are clearly defined and can be implemented by front-end developers.

[0052] 3D simulation preview of the design scheme: Integrating the Three.js engine (supporting WebGL acceleration), the specific functions include: ① Model loading: Supports importing STEP format digital models, automatically generating a 3D mesh model after loading; ② Operation functions: Rotation (mouse drag, 0-360°), zoom (scroll wheel zoom, 1-5x), cross-section viewing (mouse draws cross-section lines, cross-section color is set to red); ③ Collision detection: Setting a gap ≤ 0.5mm as interference, the interference location is highlighted in red, labeled "Interference component + gap value" (e.g., "Door and pillar interference, gap -0.3mm"); ④ Parameter labeling: Clicking on a component automatically labels "strength, weight, lifespan, carbon emissions" (labeling positions are attached to the component surface and can be manually dragged and adjusted); ⑤ Hide / show: Selecting a component name allows you to hide / show that component (e.g., hiding the outer body panel to view the internal anti-collision beam). The simulation function's engine selection, operation logic, and detection rules are clear, and the 3D preview effect can be reproduced.

[0053] 5.2 Data Output and Deployment Adaptation Data Export and Report Generation: ① Export Formats: Excel (.xlsx, including "Solution Parameter Table, Cost Breakdown Table"), PDF (Decision Report, including 6 chapters: Cover, Solution Overview, Performance Comparison, Cost Analysis, Carbon Footprint, Risk Warning); ② Report Templates: Supports administrators uploading docx templates (customizable font, logo, and chapter order), with the system automatically filling in data (e.g., the cost breakdown table is refined to "material procurement 65%, processing 25%, transportation 10%"); ③ Export Process: Select Solution → Click "Export" → Select Format → Set Save Path → Generate File (Excel generation time ≤ 10s, PDF generation time ≤ 15s). Example: The AA6061-T6 replacement solution report includes the core conclusions of "weight reduction of 15kg, carbon emission reduction of 80kg CO2, and low patent risk." The format, template, and process of this export function are clear and easy to implement.

[0054] Multiple deployment methods are supported (all are industry-standard deployment solutions, with mature and feasible technologies): (1) Local deployment: ① Hardware requirements: server CPU ≥ 16 cores, memory ≥ 64GB, storage ≥ 1TB; ② Software environment: operating system Windows Server 2019 / Linux CentOS 8, database MySQL 8.0 (master-slave replication), middleware Tomcat 9.0; ③ Integration solution: adopt RESTful API interface (supports JSON format), integrate with PDM / PLM system, data synchronization delay ≤ 10s (system synchronizes solution data to PDM, PDM synchronizes BOM data to system).

[0055] (2) Cloud deployment: ① Cloud server configuration: 4 cores 8G (supports elastic expansion to 16 cores 32G), operating system Linux Ubuntu 20.04; ② Transmission security: adopts HTTPS protocol (SSL / TLS 1.3 certificate), data backup frequency once a day (backup to cloud object storage); ③ Access support: computer (Windows 10 / 11, macOS 12+), tablet (iPadOS 15+, Android 11+), interface adaptive resolution (1920×1080 to 2048×1536).

[0056] (3) Hybrid deployment: ① Data partitioning: Sensitive data (BOM table, cost details) are stored in the local MySQL database, and non-sensitive data (material performance library, standard documents) are stored in the cloud OSS; ② Data interconnection: The local and cloud are connected via IPsec VPN. The VPN authentication method is "username + password + certificate", and the data transmission rate is ≥50MB / s.

[0057] Please see Figure 2 The AI-based intelligent decision-making method for automotive lightweighting includes the following steps: S1 Multi-Source Heterogeneous Data Standardization Acquisition and Secure Storage: A 3D model analysis tool is used to analyze the geometric topology of different versions of automotive component 3D models, extracting structural dimensions, material identifiers, and assembly constraints to generate standardized model files. A temporal neural network model is used to perform trend fitting and outlier removal on automotive material bench test data. A semi-supervised learning algorithm is used to automatically annotate unstructured documents related to automotive lightweighting and generate structured data. The standardized data is stored in a distributed data storage system, and data security management is achieved through hierarchical access control, data encryption, and transmission encryption mechanisms. S2 Lightweighting Knowledge Graph Construction: Based on a pre-trained language model, patent documents, process manuals, and industry standard texts in the field of automotive lightweighting are processed to extract four types of entities: "materials, processes, patents, and standards," as well as the relationships between these entities. A graph database is used to store these entities and relationships to construct a knowledge graph. The knowledge graph contains nodes with unique identifiers and source information, as well as relationship edges with association strength parameters. A dynamic update mechanism for the knowledge graph is established to periodically acquire the latest industry standards and update the graph content accordingly. S3 Hybrid Intelligent Algorithm-Driven Solution Generation: By inputting material properties, operating conditions, and historical performance characteristics data through a deep residual network, it predicts the performance degradation law and fatigue life of the target material under different operating conditions; based on a Bayesian optimization algorithm and constrained by preset multi-objective optimization indices, it iteratively optimizes the processing parameters of automotive parts; utilizing the Transformer architecture to input vehicle structure data and design constraints, it generates multiple lightweight design solutions including material replacement, structural adjustment, and process routes; and establishes a model iteration and update mechanism to adjust the parameters of the prediction and optimization models based on newly added sample data. S4 Multi-Objective Optimization Decision Screening: A multi-dimensional evaluation system is constructed, encompassing basic performance, economy, environmental friendliness, reliability, and compliance. Pre-defined weights are used to comprehensively score each design scheme. Schemes exceeding constraints are eliminated using a feasible region search algorithm, and the remaining schemes are validated using a particle swarm optimization algorithm. A dimensionality reduction algorithm is employed to map the multi-dimensional optimization space to a low-dimensional plane and label the Pareto optimal solution set. The impact of key variables on target indicators is analyzed based on variance decomposition to determine optimization priorities. S5 Visualization Interaction and Result Output: Generates material performance parameter comparison charts based on data visualization tools, supports parameter weight adjustment and sorting operations; realizes the 3D model display of design schemes through a 3D simulation engine, and provides model rotation, scaling and interference detection functions; supports exporting scheme parameter tables and decision analysis reports in preset formats, adaptable to local deployment, cloud deployment or hybrid deployment modes.

[0058] Finally, it should be noted that the above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Any obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. An AI-based intelligent decision-making database system for lightweight vehicles, characterized in that: Includes the following modules: The multi-source heterogeneous data acquisition module is used to standardize the processing of automotive lightweighting-related data through a reverse engineering data conversion interface, experimental data cleaning components and a dynamic data labeling system, and to manage the data through a distributed storage unit and a safety control unit. The lightweight knowledge graph construction module is used to extract entities and relationships in the field of automotive lightweighting based on the BERT text parsing model, store the constructed knowledge graph through a graph database, and configure industry standard update units to maintain the timeliness of the graph. The intelligent algorithm module integrates a material performance prediction submodule, a process parameter optimization submodule, and a scheme generation submodule. The material performance prediction submodule adopts a deep residual network architecture, the process parameter optimization submodule is based on a Bayesian optimization algorithm, and the scheme generation submodule adopts a Transformer architecture. The multi-objective optimization decision module is used to process constraints and output optimization solutions based on a multi-dimensional evaluation index system that includes basic performance, economy, environmental protection, reliability and compliance, through a feasible domain search algorithm. The visualization and interaction module provides material performance comparison, 3D preview of solutions, and data export functions, and supports local, cloud, and hybrid deployment modes.

2. The intelligent decision-making database system for lightweight vehicles based on artificial intelligence according to claim 1, characterized in that: The reverse engineering data conversion interface of the multi-source heterogeneous data acquisition module is configured with a parameter extraction algorithm based on the OpenCASCADE framework. This algorithm includes the following steps: Read the geometric topology of the 3D digital model and obtain the coordinate information of faces, edges and vertices; Match a preset feature point template, the template containing feature parameters such as sheet metal thickness threshold of 0.5-5mm, hole diameter deviation of ±0.1mm, and hole position coordinate accuracy of ±0.05mm; Automatically extract structural dimensions, material identification, and assembly constraints of components; Output a standardized digital model in STEP format.

3. The intelligent decision-making database system for lightweight vehicles based on artificial intelligence according to claim 1, characterized in that: The distributed storage unit adopts the Hadoop Distributed File System and includes one master node, one standby node and at least eight data nodes. The switching condition between the master node and the standby node is that the master node has no response time for more than 30 seconds. The single-node storage capacity of the data nodes is not less than 10TB, the data block size is 128MB and the number of replicas is 3.

4. The intelligent decision-making database system for lightweight vehicles based on artificial intelligence according to claim 1, characterized in that: The safety control unit includes: The access control component based on the RBAC model is configured with permission scopes for three roles: engineers, administrators, and suppliers. Engineers only have the right to view and edit data within their own department, administrators have the right to manage all data, and suppliers only have the right to view authorized process data. The data encryption component uses the AES-256 encryption algorithm to encrypt structured data and the SSL / TLS 1.3 protocol to encrypt unstructured data during transmission. The data masking component automatically masks data shared outside the system based on regular expressions.

5. The intelligent decision-making database system for lightweight vehicles based on artificial intelligence according to claim 1, characterized in that: The BERT text parsing model of the lightweight knowledge graph construction module adopts the bert-base-chinese pre-trained model, which has been fine-tuned and trained with 8,000 labeled documents, including patent documents, process manuals and industry standards. The model's entity extraction F1 score is no less than 0.92 and the relationship recognition accuracy is no less than 0.

90. The extracted core relationships include the relationship between process and applicable materials, and the relationship between materials and prohibited or restricted substances.

6. The intelligent decision-making database system for lightweight vehicles based on artificial intelligence according to claim 1, characterized in that: The material performance prediction submodule uses a ResNet-50 deep residual network, which contains 50 3×3 convolutional layers and 16 residual blocks. The input features are 28-dimensional, including 8 basic material properties, 6 operating conditions, and 14 historical performance parameters. The output dimension is 2, corresponding to the strength decay value and the number of lifetime cycles, respectively. The training dataset of the submodule contains more than 200,000 sets of test data covering 32 types of automotive materials, with the test operating temperature range from -40℃ to 120℃.

7. The intelligent decision-making database system for lightweight vehicles based on artificial intelligence according to claim 1, characterized in that: The Bayesian optimization algorithm of the process parameter optimization submodule uses the RBF kernel function with a noise coefficient of 0.

1. The training data of this submodule includes more than 20,000 sets of historical process data, which includes the parameter range of thermoforming temperature 500-900℃ and pressure 10-50MPa. The optimization objective is to maximize the yield weight of 0.6 and minimize the forming time weight of 0.

4. The iterative process includes parameter initialization, objective function value calculation, Gaussian process model update and parameter selection steps, with 30 iterations.

8. The intelligent decision-making database system for lightweight vehicles based on artificial intelligence according to claim 1, characterized in that: The multi-dimensional evaluation index system of the multi-objective optimization decision module includes: The basic performance index has a weight of 0.3 and includes a weight score and a strength score. The weight score is calculated by subtracting the actual value from the upper limit value and dividing by the upper limit value, then multiplying by 30. The strength score is calculated by dividing the actual value by the required value and multiplying by 30. The economic indicator has a weight of 0.25 and includes material cost score and process cost score. The material cost score is calculated by subtracting the actual value from the budget and dividing by the budget, then multiplying by 25. The process cost score is calculated by subtracting the actual cost from the benchmark cost and dividing by the benchmark cost, then multiplying by 25. The environmental performance index has a weight of 0.

2. The carbon emission score is calculated by subtracting the actual value from the benchmark value and dividing by the benchmark value by 20 points. The benchmark value is the carbon emission value of the original steel scheme. The reliability index has a weight of 0.15, and the fatigue life score is calculated by dividing the actual value by the required value and multiplying by 15. The compliance indicator has a weight of 0.1 and includes a patent risk score of 5 points for low risk, 3 points for medium risk, and 0 points for high risk, as well as a compliance score of 5 points for compliance and 0 points for non-compliance for prohibited and restricted substances.

9. The intelligent decision-making database system for lightweight vehicles based on artificial intelligence according to claim 1, characterized in that: The 3D simulation preview unit of the visualization interaction module integrates the Three.js engine and supports the following functions: Import the STEP format digital model and generate a 3D mesh model; The model can be rotated 360 degrees by dragging with the mouse, zoomed from 1 to 5 times by scroll wheel, and viewed by section lines; Collision detection is performed on areas with gaps ≤0.5mm, and these areas are highlighted in red. It can label the strength and weight parameters of parts and supports dragging and adjusting the position of the parameter labels.

10. An AI-based intelligent decision-making method for lightweight vehicles, wherein the method is applied to the AI-based intelligent decision-making database system for lightweight vehicles as described in any one of claims 1 to 9, characterized in that: The method includes the following steps: S1 Multi-Source Heterogeneous Data Standardization Acquisition and Secure Storage: A 3D model analysis tool is used to analyze the geometric topology of different versions of automotive component 3D models, extracting structural dimensions, material identifiers, and assembly constraints to generate standardized model files. A temporal neural network model is used to perform trend fitting and outlier removal on automotive material bench test data. A semi-supervised learning algorithm is used to automatically annotate unstructured documents related to automotive lightweighting and generate structured data. The standardized data is stored in a distributed data storage system, and data security management is achieved through hierarchical access control, data encryption, and transmission encryption mechanisms. S2 Lightweight Knowledge Graph Construction: Based on a pre-trained language model, patent documents, process manuals, and industry standard texts in the field of automotive lightweighting are processed to extract four types of entities: materials, processes, patents, and standards, as well as the relationships between these entities. A graph database is used to store these entities and relationships to construct a knowledge graph. The knowledge graph contains nodes with unique identifiers and source information, as well as relationship edges with association strength parameters. A dynamic update mechanism for the knowledge graph is established to regularly acquire the latest industry standards and update the graph content accordingly. S3 Hybrid Intelligent Algorithm-Driven Solution Generation: By inputting material properties, operating conditions, and historical performance characteristics data through a deep residual network, it predicts the performance degradation law and fatigue life of the target material under different operating conditions; based on a Bayesian optimization algorithm and constrained by preset multi-objective optimization indices, it iteratively optimizes the processing parameters of automotive parts; utilizing the Transformer architecture to input vehicle structure data and design constraints, it generates multiple lightweight design solutions including material replacement, structural adjustment, and process routes; and establishes a model iteration and update mechanism to adjust the parameters of the prediction and optimization models based on newly added sample data. S4 Multi-Objective Optimization Decision Screening: A multi-dimensional evaluation system is constructed, encompassing basic performance, economy, environmental friendliness, reliability, and compliance. Pre-defined weights are used to comprehensively score each design scheme. Schemes exceeding constraints are eliminated using a feasible region search algorithm, and the remaining schemes are validated using a particle swarm optimization algorithm. A dimensionality reduction algorithm is employed to map the multi-dimensional optimization space to a low-dimensional plane and label the Pareto optimal solution set. The impact of key variables on target indicators is analyzed based on variance decomposition to determine optimization priorities. S5 Visualization Interaction and Result Output: Generates material performance parameter comparison charts based on data visualization tools, supports parameter weight adjustment and sorting operations; realizes the 3D model display of design schemes through a 3D simulation engine, and provides model rotation, scaling and interference detection functions; supports exporting scheme parameter tables and decision analysis reports in preset formats, adaptable to local deployment, cloud deployment or hybrid deployment modes.

Images