Intelligent manufacturing line optimization simulation platform based on digital twinning
By combining multi-source intelligent data acquisition, dynamic twin modeling, and collaborative optimization engine modules, and utilizing transfer learning and federated learning algorithms, the problem of rapid adaptation of digital twin models was solved, enabling real-time optimization and efficient simulation of the production line, thereby improving the scientific nature of production decisions and optimization efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- LEGER TECH SERVICES LTD
- Filing Date
- 2025-10-25
- Publication Date
- 2026-06-19
Smart Images
Figure CN122243308A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent manufacturing technology, specifically to an intelligent manufacturing production line optimization simulation platform based on digital twins. Background Technology
[0002] In the field of intelligent manufacturing, production line optimization simulation technology based on digital twins is a technical means to improve production line efficiency and the scientific nature of decision-making by constructing a digital model that is precisely mapped to the physical production line and integrating production line operation data in real time to simulate, analyze and optimize the production process.
[0003] In existing technologies, the application of digital twin technology in manufacturing production lines mainly involves collecting information such as equipment operating parameters and production process data to construct static or semi-dynamic digital models for offline simulation analysis of the production line. For example, it can simulate the capacity under different production plans or conduct retrospective analysis of equipment failures through historical data. Some technologies combine simple machine learning algorithms to periodically update the model parameters.
[0004] However, in existing technologies, digital twin models are difficult to adapt quickly to the characteristics of new production lines, and model updates rely on centralized data processing, which cannot achieve efficient iteration based on multi-dimensional real-time data. This results in insufficient dynamic mapping accuracy between digital models and physical production lines, making it difficult to meet the demand for high-fidelity models for real-time production line optimization. In view of this, we propose a digital twin-based intelligent manufacturing production line optimization simulation platform. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention provides a digital twin-based intelligent manufacturing production line optimization simulation platform. This platform solves the problems in existing technologies where digital twin models are difficult to quickly adapt to the characteristics of new production lines, and where model updates rely on centralized data processing, making it impossible to achieve efficient iteration based on multi-dimensional real-time data.
[0006] To achieve the above objectives, the present invention provides the following technical solution: a digital twin-based intelligent manufacturing production line optimization simulation platform, comprising the following modules:
[0007] The multi-source intelligent acquisition module collects pre-processed production equipment operating parameters, production process data, workshop environmental parameters, personnel operation data, and material flow information. The data includes timestamps and spatial coordinate identifiers.
[0008] The dynamic twin modeling module receives data output from the multi-source intelligent acquisition module, uses transfer learning algorithms to initialize multi-dimensional models of equipment, processes, logistics, and personnel, achieves real-time iterative updates of the model through federated learning mechanisms, and generates a visual model that dynamically maps to the physical production line by combining 3D modeling technology.
[0009] The collaborative optimization engine module receives real-time model data output by the dynamic twin modeling module and uses a deep reinforcement learning algorithm to construct a global optimization model for the production line. This algorithm forms a parameter interaction closed loop with the federated learning mechanism of the dynamic twin modeling module, and synchronously executes equipment parameter optimization, production path planning, and resource scheduling optimization.
[0010] The virtual-real interaction control module interacts bidirectionally with the collaborative optimization engine module, providing a three-dimensional visualization operation interface that allows users to configure optimization constraints, trigger simulation scenarios, and recall historical optimization schemes.
[0011] The knowledge graph storage module receives model iteration data from the dynamic twin modeling module and optimization result data from the collaborative optimization engine module to construct a production line optimization knowledge graph, which includes equipment characteristic parameters, process optimization rules, and solutions to typical problems.
[0012] Preferably, the multi-source intelligent acquisition module deploys an industrial-grade sensor array and an intelligent data acquisition terminal, transmits data through a time-sensitive network, and performs outlier removal, format standardization, and spatiotemporal alignment operations on the acquired data through edge nodes. The data sampling interval switches between 0.1 seconds and 10 seconds depending on the production line conditions.
[0013] Preferably, the multi-source intelligent acquisition module also acquires equipment energy consumption data, tool life data, and product quality inspection data, and stores them by classification through data tags, which include data source identifiers, acquisition time identifiers, and data type identifiers.
[0014] Preferably, the transfer learning algorithm of the dynamic twin modeling module extracts common feature parameters from the historical production line dataset, adjusts the model adaptation coefficients through real-time data of the newly added production line, and uses point cloud scanning technology to restore the geometric features of the equipment in 3D modeling. The model update frequency is synchronized with the data transmission frequency of the multi-source intelligent acquisition module.
[0015] Preferably, the federated learning mechanism of the dynamic twin modeling module sets model parameter aggregation rules, assigns weight coefficients according to the importance of each dimension of the model, and the weight coefficients are manually adjusted through the virtual-real interaction control module, with an adjustment range of 0.1 to 0.9.
[0016] Preferably, the deep reinforcement learning algorithm of the collaborative optimization engine module includes a state perception layer, a decision execution layer, and a feedback adjustment layer. The state perception layer parses the model state vector output by the dynamic twin modeling module, the decision execution layer generates equipment control parameters and production scheduling instructions, and the feedback adjustment layer sends the optimization effect parameters back to the dynamic twin modeling module.
[0017] Preferably, the collaborative optimization engine module sets optimization iteration termination conditions, including a maximum iteration number threshold, an optimization target error threshold, and a parameter convergence threshold. The threshold values are configured through the virtual-real interaction control module, and the configuration range is ±50% of the preset benchmark value.
[0018] Preferably, the three-dimensional visualization interface of the virtual-real interaction control module supports 1:1 scene rendering and includes a parameter input area, a scene display area, and a result output area. The parameter input area provides numerical input boxes, option buttons, and sliding controls, while the result output area displays the production line operation indicators before and after optimization in the form of charts.
[0019] Preferably, the knowledge graph storage module adopts a distributed graph database architecture, and the stored data includes equipment model parameters, process standard values, historical optimization parameters, and fault handling records. The data update frequency is consistent with the model iteration frequency of the dynamic twin modeling module.
[0020] Preferably, the knowledge graph storage module establishes a data query channel with the virtual-real interaction control module, supporting users to search for equipment parameters, process rules, and optimization cases through keywords. The search results are displayed in the form of structured tables and association graphs.
[0021] This invention provides a digital twin-based intelligent manufacturing production line optimization simulation platform. It offers the following advantages:
[0022] 1. This invention uses the transfer learning algorithm of the dynamic twin modeling module to extract common feature parameters from historical production line datasets, and adjusts the model adaptation coefficients by combining them with real-time data from newly added production lines to quickly build an initial model. Then, it uses a federated learning mechanism to aggregate model parameters from various dimensions to achieve real-time iterative updates of the model, enabling the digital twin model to accurately map the state of the physical production line, providing reliable input for the collaborative optimization engine, and improving the accuracy and timeliness of production line simulation.
[0023] 2. This invention utilizes a deep reinforcement learning algorithm in the collaborative optimization engine module. The state vector of the model is analyzed by the state perception layer, the optimization action is generated by the decision execution layer, and the network parameters are updated by the feedback adjustment layer. This forms a parameter interaction closed loop with the dynamic twin modeling module, which can simultaneously optimize equipment parameters, production paths, and resource scheduling, thereby achieving efficient global optimization of the production line and enhancing the scientific nature of production decisions.
[0024] 3. This invention integrates iterative data from dynamic twin models with collaborative optimization results through a knowledge graph storage module, constructing a knowledge graph containing equipment parameters, process rules, etc., and linking with the virtual-real interaction control module to support data retrieval. This process transforms optimization experience into knowledge, providing a reference for subsequent production line optimization and improving optimization efficiency and sustainability. Attached Figure Description
[0025] Figure 1This is a flowchart of a digital twin-based intelligent manufacturing production line optimization simulation platform system. Detailed Implementation
[0026] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0027] Example:
[0028] Please see the appendix Figure 1 This invention provides a digital twin-based intelligent manufacturing production line optimization simulation platform, comprising the following modules:
[0029] The multi-source intelligent acquisition module collects pre-processed production equipment operating parameters, production process data, workshop environmental parameters, personnel operation data, and material flow information. The data includes timestamps and spatial coordinate identifiers.
[0030] The multi-source intelligent acquisition module deploys an industrial-grade sensor array and an intelligent data acquisition terminal. It transmits data through a time-sensitive network. The acquired data undergoes outlier removal, format standardization, and spatiotemporal alignment at edge nodes. The data sampling interval switches between 0.1 seconds and 10 seconds depending on the production line conditions.
[0031] The multi-source intelligent acquisition module also collects equipment energy consumption data, tool life data, and product quality inspection data, and stores them by data tags. The tags include data source identifier, acquisition time identifier, and data type identifier.
[0032] The dynamic twin modeling module receives data output from the multi-source intelligent acquisition module, uses transfer learning algorithms to initialize multi-dimensional models of equipment, processes, logistics, and personnel, achieves real-time iterative updates of the model through federated learning mechanisms, and generates a visual model that dynamically maps to the physical production line by combining 3D modeling technology.
[0033] The transfer learning algorithm of the dynamic twin modeling module extracts common feature parameters from historical production line datasets, adjusts the model adaptation coefficients using real-time data from newly added production lines, and uses point cloud scanning technology to restore the geometric features of the equipment for 3D modeling. The model update frequency is synchronized with the data transmission frequency of the multi-source intelligent acquisition module, and includes the following algorithm steps:
[0034] I. Initialization and preparation of the dynamic twin model, starting from historical production line datasets. Extracting general feature parameters ,in This is a collection of various data accumulated during the historical operation of the production line, including equipment parameters, process data, etc. To identify common characteristics in these data, such as general operating patterns of equipment and typical process parameter ranges, and to acquire initial real-time data for new production lines. , This refers to the pre-processed data collected in the initial stage for the production line that needs to be modeled.
[0035] II. Constructing an initial model using transfer learning algorithms, and generating a multi-dimensional initial model through formulas. :
[0036]
[0037] in:
[0038] TL stands for Transfer Learning Function, which is responsible for transferring general features to the new model. The initial model adaptation coefficient is used to adjust the degree of matching between general features and new production line data, so that the initial model can initially adapt to the characteristics of the new production line. The initial model includes a basic model framework with four dimensions: equipment, process, logistics, and personnel.
[0039] 3. Initial construction of the 3D model: Point cloud scanning technology is used to scan the physical production line equipment to obtain point cloud data of the equipment surface. , This is a set of three-dimensional coordinate information containing a large number of points on the surface of the device, and is modeled using the 3D Modeling Algorithm 3DMod ( Generate a 3D model that matches the geometric features of the physical device. , As the initial 3D visualization model, it is compared with the multi-dimensional initial model obtained in step 2. By linking them together, a preliminary dynamic twin visualization model is formed;
[0040] The federated learning mechanism of the dynamic twin modeling module sets model parameter aggregation rules and assigns weight coefficients according to the importance of each dimension of the model. The weight coefficients are manually adjusted through the virtual-real interaction control module, with an adjustment range of 0.1 to 0.9, and include the following algorithm steps:
[0041] I. Set initial parameters for federated learning and determine the initial weight coefficients for the four dimensions of the model: equipment, process, logistics, and personnel. ( (corresponding to four dimensions respectively) The initial values are set based on the typical importance of each dimension in the production line, and are within the range of 0.1 to 0.9. The model update frequency is also configured. ,make Data transmission frequency with multi-source intelligent acquisition module Maintain consistency, that is This ensures that model updates are synchronized with the data collection schedule.
[0042] II. Real-time reception and processing of new data: The multi-source intelligent acquisition module continuously transmits real-time data from newly added production lines. ( (Representing time), the dynamic twin modeling module receives... ,in It contains various real-time operational data of the production line at that moment, which is used to drive model updates.
[0043] III. Each dimension of the model is updated independently, based on... The models for the four dimensions of equipment, process, logistics, and personnel were updated independently to obtain... Updated model at each time dimension ( ), Reflects the corresponding dimension in The latest status at any given time is updated by incorporating characteristic algorithms from various dimensions, such as combining equipment models with the changing patterns of equipment operating parameters, and combining process models with real-time process execution data.
[0044] IV. Federated learning aggregates and updates the model, aggregating the models across various dimensions according to a formula:
[0045]
[0046] get Global dynamic twin model at time step ,in The weight coefficients for each dimension currently set can be manually adjusted through the virtual-real interaction control module. After adjustment, they will still be within the range of 0.1 to 0.9. Through this aggregation process, the updated information of each dimension is integrated to form a global model that reflects the overall status of the production line.
[0047] V. Real-time updates of the 3D model, based on The device status data and new point cloud data (such as data collected when the device location or shape changes) are updated using the 3D model update algorithm Update3D. , Update the 3D visualization model to obtain... 3D model of time , for The point cloud data at any given time (if updated) keeps the 3D model consistent with the global dynamic twin model, enabling dynamic mapping between the visualization model and the physical production line.
[0048] 6. Output the updated model to the collaborative optimization engine. Global dynamic twin model at time step The data is transmitted to the collaborative optimization engine module as input to the state-aware layer in the deep reinforcement learning algorithm, i.e., through the process described above. Enc Formula generates state vector This enables algorithmic linkage between the dynamic twin modeling module and the collaborative optimization engine module, providing a foundation for subsequent optimization decisions.
[0049] 7. Repeat steps two through six, according to the set update frequency. It continuously receives new data, updates models in various dimensions, aggregates the global model, updates the 3D model and outputs it, so that the dynamic twin model always keeps dynamically synchronized with the physical production line, while continuously providing real-time model data to the collaborative optimization engine module to support the continuous optimization process;
[0050] The collaborative optimization engine module receives real-time model data output from the dynamic twin modeling module and constructs a global optimization model for the production line using a deep reinforcement learning algorithm. This algorithm forms a parameter interaction closed loop with the federated learning mechanism of the dynamic twin modeling module, synchronously executing equipment parameter optimization, production path planning, and resource scheduling optimization. The deep reinforcement learning algorithm of the collaborative optimization engine module includes a state perception layer, a decision execution layer, and a feedback adjustment layer. The state perception layer parses the model state vector output by the dynamic twin modeling module, the decision execution layer generates equipment control parameters and production scheduling instructions, and the feedback adjustment layer sends the optimization effect parameters back to the dynamic twin modeling module. The collaborative optimization engine module sets optimization iteration termination conditions, including a maximum iteration threshold, an optimization target error threshold, and a parameter convergence threshold. The threshold values are configured through the virtual-real interaction control module, with a configuration range of ±50% of a preset baseline value. The algorithm includes the following steps:
[0051] I. Initialize the dynamic twin model and collaborative optimization engine parameters. The dynamic twin modeling module, based on the preprocessed data from the multi-source intelligent acquisition module, constructs an initial production line model through transfer learning algorithms. Its core formula is:
[0052]
[0053] in:
[0054] This represents the initially constructed digital twin model;
[0055] TL stands for transfer learning function;
[0056] This is a historical production line dataset;
[0057] This is newly collected data for the current production line;
[0058] This step quickly builds an initial model framework adapted to the current production line by transferring common features from historical data; the collaborative optimization engine module initializes deep reinforcement learning parameters and sets policy network parameters. and value network parameters The policy network is used to generate optimized actions, and the value network is used to evaluate the value of the actions.
[0059] II. The dynamic twin model is updated in real time. The dynamic twin modeling module uses a federated learning mechanism to receive real-time data from the multi-source intelligent acquisition module and updates the model through formulas:
[0060]
[0061] in:
[0062] The updated model at time t;
[0063] FL represents the federated learning aggregation function;
[0064] for A model of time;
[0065] This refers to the newly acquired data at time t;
[0066] These are the weighting coefficients for each data source, set according to the importance of the data.
[0067] This step enables the model to iterate in real time while maintaining data privacy, ensuring that the model is consistent with the status of the physical production line.
[0068] III. Collaborative Optimization Engine State Awareness: The state awareness layer of the deep reinforcement learning algorithm analyzes the model state vector output by the dynamic twin modeling module, using the formula... model Encoded as a state vector Where Enc is the state coding function. This step, which includes key information such as production line equipment operating parameters, material locations, and personnel status, transforms complex model data into state representations that can be processed by reinforcement learning.
[0069] IV. Generating optimized decisions; the decision execution layer is based on state vectors. Through policy network Generate optimization actions The formula is ,in For policy network parameters, This step includes equipment control parameters, production path instructions, and resource scheduling schemes. Based on the current production line status, this step outputs specific optimized operation instructions.
[0070] V. Execute optimization decisions and obtain feedback; the dynamic twin modeling module receives optimization actions. After the simulation is executed, a new model state is output. The collaborative feedback adjustment layer calculates immediate rewards. The formula is:
[0071]
[0072] in:
[0073] For the reward function;
[0074] To perform the action The new state afterwards;
[0075] The reward value is set based on optimization goals such as improving production efficiency and reducing energy consumption. This step evaluates the effectiveness of the optimization actions through model simulation.
[0076] VI. Update the reinforcement learning network parameters using the temporal difference algorithm, with the following formula:
[0077]
[0078] in:
[0079] The learning rate;
[0080] This refers to timing difference error;
[0081] This is a discount factor (ranging from 0 to 1, used to balance immediate rewards and future rewards).
[0082] The state value output by the value network;
[0083] Simultaneously, the policy network is updated using the policy gradient algorithm, with the following formula:
[0084]
[0085] in:
[0086] The learning rate is the policy rate.
[0087] The action value function;
[0088] This step continuously optimizes network parameters and improves decision-making quality through feedback and rewards.
[0089] 7. To determine whether the optimization iteration has terminated, the collaborative optimization engine module checks whether the termination condition is met. If the maximum number of iterations reaches a preset threshold... Optimize target error (in The current target value, To achieve the optimal value, (for error threshold) or parameter convergence ( If the threshold for convergence is reached, the iteration stops and the optimal solution is output; otherwise, return to step 2 to continue iterating. This step ensures that the optimization process terminates efficiently within a reasonable range, avoiding over-computation or insufficient optimization.
[0090] 8. Store the optimization results in the knowledge graph. The knowledge graph storage module receives the final optimization parameters and dynamic twin model data, and then stores them using the formula... Update the knowledge graph, in which For knowledge graphs, This is the optimal state. For optimal action, To achieve the best reward, this step will transform optimization experience into knowledge, providing a reference for subsequent optimization tasks;
[0091] The virtual-real interaction control module interacts bidirectionally with the collaborative optimization engine module, providing a three-dimensional visualization operation interface that allows users to configure optimization constraints, trigger simulation scenarios, and recall historical optimization schemes.
[0092] The three-dimensional visualization interface of the virtual-real interaction control module supports 1:1 scene rendering and includes a parameter input area, a scene display area, and a result output area. The parameter input area provides numerical input boxes, option buttons, and sliding controls, while the result output area displays the production line operation indicators before and after optimization in the form of charts.
[0093] The knowledge graph storage module receives model iteration data from the dynamic twin modeling module and optimization result data from the collaborative optimization engine module to construct a production line optimization knowledge graph, which includes equipment characteristic parameters, process optimization rules, and solutions to typical problems.
[0094] The knowledge graph storage module adopts a distributed graph database architecture. The stored data includes equipment model parameters, process standard values, historical optimization parameters, and fault handling records. The data update frequency is consistent with the model iteration frequency of the dynamic twin modeling module.
[0095] The knowledge graph storage module establishes a data query channel with the virtual-real interaction control module, allowing users to search for equipment parameters, process rules, and optimization cases using keywords. The search results are displayed in the form of structured tables and relational graphs.
[0096] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A digital twin-based intelligent manufacturing production line optimization simulation platform, characterized in that, Includes the following modules: The multi-source intelligent acquisition module collects pre-processed production equipment operating parameters, production process data, workshop environmental parameters, personnel operation data, and material flow information. The data includes timestamps and spatial coordinate identifiers. The dynamic twin modeling module receives data output from the multi-source intelligent acquisition module, uses transfer learning algorithms to initialize multi-dimensional models of equipment, processes, logistics, and personnel, achieves real-time iterative updates of the model through federated learning mechanisms, and generates a visual model that dynamically maps to the physical production line by combining 3D modeling technology. The collaborative optimization engine module receives real-time model data output by the dynamic twin modeling module and uses a deep reinforcement learning algorithm to construct a global optimization model for the production line. This algorithm forms a parameter interaction closed loop with the federated learning mechanism of the dynamic twin modeling module, and synchronously executes equipment parameter optimization, production path planning, and resource scheduling optimization. The virtual-real interaction control module interacts bidirectionally with the collaborative optimization engine module, providing a three-dimensional visualization operation interface that allows users to configure optimization constraints, trigger simulation scenarios, and recall historical optimization schemes. The knowledge graph storage module receives model iteration data from the dynamic twin modeling module and optimization result data from the collaborative optimization engine module to construct a production line optimization knowledge graph, which includes equipment characteristic parameters, process optimization rules, and solutions to typical problems.
2. The intelligent manufacturing production line optimization simulation platform based on digital twins according to claim 1, characterized in that, The multi-source intelligent acquisition module deploys an industrial-grade sensor array and an intelligent data acquisition terminal. It transmits data through a time-sensitive network. The acquired data undergoes outlier removal, format standardization, and spatiotemporal alignment at edge nodes. The data sampling interval switches between 0.1 seconds and 10 seconds depending on the production line conditions.
3. The intelligent manufacturing production line optimization simulation platform based on digital twins according to claim 1, characterized in that, The multi-source intelligent acquisition module also collects equipment energy consumption data, tool life data, and product quality inspection data, and stores them by classification through data tags. The tags include data source identifier, acquisition time identifier, and data type identifier.
4. The intelligent manufacturing production line optimization simulation platform based on digital twins according to claim 1, characterized in that, The transfer learning algorithm of the dynamic twin modeling module extracts common feature parameters from historical production line datasets, adjusts the model adaptation coefficients using real-time data from newly added production lines, and uses point cloud scanning technology to restore the geometric features of the equipment in 3D modeling. The model update frequency is synchronized with the data transmission frequency of the multi-source intelligent acquisition module.
5. The intelligent manufacturing production line optimization simulation platform based on digital twins according to claim 1, characterized in that, The federated learning mechanism of the dynamic twin modeling module sets the aggregation rules for model parameters and assigns weight coefficients according to the importance of each dimension of the model. The weight coefficients are manually adjusted through the virtual-real interaction control module, and the adjustment range is from 0.1 to 0.
9.
6. The intelligent manufacturing production line optimization simulation platform based on digital twins according to claim 1, characterized in that, The deep reinforcement learning algorithm of the collaborative optimization engine module includes a state perception layer, a decision execution layer, and a feedback adjustment layer. The state perception layer parses the model state vector output by the dynamic twin modeling module, the decision execution layer generates equipment control parameters and production scheduling instructions, and the feedback adjustment layer sends the optimization effect parameters back to the dynamic twin modeling module.
7. The intelligent manufacturing production line optimization simulation platform based on digital twins according to claim 1, characterized in that, The collaborative optimization engine module sets optimization iteration termination conditions, including a maximum iteration number threshold, an optimization target error threshold, and a parameter convergence threshold. The threshold values are configured through the virtual-real interaction control module, and the configuration range is ±50% of the preset benchmark value.
8. The intelligent manufacturing production line optimization simulation platform based on digital twins according to claim 1, characterized in that, The three-dimensional visualization interface of the virtual-real interaction control module supports 1:1 scene rendering and includes a parameter input area, a scene display area, and a result output area. The parameter input area provides numerical input boxes, option buttons, and sliding controls, while the result output area displays the production line operation indicators before and after optimization in the form of charts.
9. The intelligent manufacturing production line optimization simulation platform based on digital twins according to claim 1, characterized in that, The knowledge graph storage module adopts a distributed graph database architecture. The stored data includes equipment model parameters, process standard values, historical optimization parameters, and fault handling records. The data update frequency is consistent with the model iteration frequency of the dynamic twin modeling module.
10. The intelligent manufacturing production line optimization simulation platform based on digital twins according to claim 1, characterized in that, The knowledge graph storage module establishes a data query channel with the virtual-real interaction control module, allowing users to search for equipment parameters, process rules, and optimization cases using keywords. The search results are displayed in the form of structured tables and relational graphs.