Hybrid Forecasting of Technical System State Under Acting Forces
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computer-aided simulations for technical systems are limited due to complex physical models' computational intensity and data-based models' need for extensive measurement data, restricting their use as forecasting and diagnostic tools during real machine operation.
Innovation Solution
A method for configuring a forecasting device that uses an approximator to predict physical behavior based on acting forces, where configuration parameters are iteratively adjusted to minimize deviations between predicted and measured system states, allowing for more accurate and efficient modeling with reduced data requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physical models are used for computer-aided simulations, then forecasting accuracy is improved, but computational complexity and resource requirements increase
Solution Approach 1:
The patent segments the modeling task into two parts: a simplified data-based model for routine forecasting and a more complex physical model for validation and critical scenarios. This segmentation allows the system to achieve accurate forecasting when needed while avoiding unnecessary computational complexity for routine operations.
Solution Approach 2:
The patent changes the parameters of the model by using dimensionless numbers and simplified physical relationships in certain conditions, allowing the model to maintain accuracy while reducing computational complexity. The model adapts its complexity based on the specific operating conditions and requirements.
2Productivity
If data-based models are used for computer-aided simulations, then computational efficiency is improved, but the amount of measurement data required increases
Solution Approach 1:
The patent introduces physical laws and dimensionless numbers as intermediaries between raw measurement data and the forecasting model. This intermediary layer allows the model to work with reduced data requirements by leveraging established physical relationships rather than requiring extensive training data for all scenarios.
Solution Approach 2:
The patent performs preliminary actions by pre-processing measurement data to extract relevant features and relationships that conform to physical laws. This preliminary processing reduces the amount of raw data needed while maintaining the model's ability to make accurate forecasts.
3Measurement precision
If complex physical models are used, then model accuracy is improved, but the time required for simulation increases
Solution Approach 1:
The patent makes the model dynamic by adjusting its complexity based on real-time conditions. The system uses a hybrid approach that switches between simplified and detailed physical models depending on the specific forecasting needs, allowing accurate results without always incurring the full computational time cost of complex models.
4Ease of manufacture
If data-based models are used, then ease of implementation is improved, but extrapolation capabilities worsen
Solution Approach 1:
The patent replaces pure data-based statistical methods with a hybrid approach that incorporates physical laws and dimensionless analysis. This substitution maintains the ease of implementation of data-based models while significantly improving extrapolation capabilities by grounding the model in fundamental physical principles that hold across different operating conditions.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
The invention relates to a forecasting device (100) for a technical system for forecasting a system state of the technical system at a future time and a method for configuring the forecasting device. The forecasting device (100) is configured to forecast a physical behavior of the technical system as a function of a force acting on the technical system, starting from a first read-in system state, wherein the acting force is determined by an approximator (AP, NN), and to output a resulting system state at a given time, wherein configuration parameters of the approximator are set using measured system states in such a way as to minimize a deviation of a resulting system state from a measured system state following the first system state at the given time.