Hybrid Digital Twin Control With Physics Blocks and Limited Sensor Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing digital twin technologies face challenges in achieving high-quality performance, especially when there is limited sensor data available, as they struggle to accurately model the behavior of entities in communications networks.
Innovation Solution
The implementation of a computer-implemented method using a digital twin that combines at least one learning block and one physics block. The learning block utilizes supervised learning and backpropagation to process sensor data, while the physics block encodes rules, enabling the digital twin to generate outputs that effectively control the entity, even with limited sensor data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional digital twin modeling is used with limited sensor data, then the model accuracy deteriorates, but using more sensors increases cost and complexity
Solution Approach 1:
The patent combines physics-based models (encoding domain knowledge and rules) with machine learning models into a hybrid digital twin architecture. This merging allows the system to leverage both the interpretability and constraints of physics models and the adaptive learning capabilities of ML models, achieving high accuracy even with limited sensor data by compensating through the integrated approach rather than relying solely on increased data quantity
Solution Approach 2:
The physics-based component acts as an intermediary that bridges the gap between limited sensor data and accurate model predictions. By encoding domain knowledge and physical constraints, it provides a framework that guides the machine learning component, enabling the digital twin to make accurate predictions without requiring extensive sensor data
2Adaptability or versatility
If purely data-driven machine learning is used, then adaptability to new patterns improves, but reliability and interpretability deteriorate
Solution Approach 1:
The patent merges data-driven machine learning components with physics-based modeling components in a hybrid architecture. The machine learning part provides adaptability to new patterns and data-driven insights, while the physics-based part ensures reliability through domain knowledge constraints and physical laws. This combination allows the digital twin to maintain both adaptability and reliability simultaneously
Solution Approach 2:
The system dynamically adjusts the weighting and influence of different components (physics-based vs. data-driven) based on the specific task and available data. This parameter adjustment allows the digital twin to optimize between adaptability and reliability depending on the operational context, using more physics-based reasoning when reliability is critical and more data-driven approaches when adaptability is needed
3Measurement precision
If complex hybrid models combining physics and learning are used, then model accuracy improves, but device complexity increases
Solution Approach 1:
The patent segments the digital twin into distinct modular components: physics-based modules encoding domain knowledge and rules, machine learning modules for pattern recognition, and integration layers that coordinate them. This segmentation allows each component to be developed, validated, and maintained independently while contributing to the overall high accuracy of the digital twin, managing complexity through modular architecture
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A digital twin is described for controlling entities such as software defined communications networks. Sensor data is observed from an entity and input to a digital twin of the entity, the digital twin comprising at least one learning block, and at least one physics block encoding a rule. The sensor data is processed through the learning block and the physics block such that the data follows a path through the digital twin through the blocks and along one or more connections between the blocks to generate an output. Automated control of the entity may be triggered in dependence on the generated output.