Digital Twin Model Selection via Dimensional Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Digital twin simulations face challenges with machine learning models when real-world data differs from training data, leading to inaccuracies, and scenario sets suffer from the 'curse of dimensionality,' making high-speed search for optimal models difficult, especially when dealing with scenarios of large dimensions.
Innovation Solution
A system determines model classes based on physical or digital asset data factors and thresholds, generates scenario models within specified outcome value ranges, and selects an optimal scenario model for digital twin simulation, enabling high-speed searches even with large dimensions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a scenario set with many dimensions is used to improve accuracy, then the accuracy of the digital twin simulation is improved, but the search range becomes wide and search time increases
Solution Approach 1:
The patent segments the scenario model into multiple dimensions (first dimension, second dimension, third dimension, etc.), each representing different aspects of the scenario. By dividing the complex high-dimensional search space into separate dimensional components, the system can process and search each dimension independently, reducing the overall search complexity while maintaining comprehensive coverage of the scenario space.
Solution Approach 2:
The patent transforms the high-dimensional scenario search problem into a multi-dimensional structured search by organizing scenarios along different dimensional axes. This dimensional organization allows the system to navigate the scenario space more efficiently by searching along each dimension separately rather than exhaustively searching the entire high-dimensional space at once.
2Adaptability or versatility
If machine learning models are used with real-world data, then the system can process actual operational data, but the estimated state may be significantly different from the actual state when data differs from training data
Solution Approach 1:
The patent performs preliminary actions by pre-defining multiple scenario models that represent different possible states and conditions before actual operation. These pre-prepared scenario models cover a range of potential situations, allowing the system to quickly match actual real-world data against predefined scenarios rather than relying solely on machine learning inference, thereby improving reliability when real-world data differs from training data.
Solution Approach 2:
The patent creates multiple copies of scenario models representing different dimensional variations of the same scenario. By having multiple copied scenario models with different dimensional characteristics, the system can select the most appropriate copy that matches the actual operational conditions, improving adaptability while maintaining accuracy through model selection rather than pure machine learning prediction.
Data Source
AI summary
A system determines a plurality of model classes of a scenario model based on: different factors with respect to a scenario model for a digital twin simulator, the factors being specified from physical or digital asset; and a result of comparison between a value for the factor and a threshold of each factor. For each of the model classes and for each outcome with respect to the scenario model, the system receives, from a user, an outcome value range that is a range of a value of the each outcome and is a range of a value based on heuristics. The system prepares, for each model class, a scenario model having an outcome value belonging to the outcome value range received for the each model class. The system selects an optimal scenario model from among the scenario models prepared for respective ones of the model classes.