Production Plant Control Using Uncertainty-Guided Design Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current design optimization methods for complex products, such as robots and turbines, require extensive computational resources due to high simulation demands, often relying on inaccurate surrogate models that consume significant computational effort.
Innovation Solution
A machine learning module is trained to generate predictive performance signals and uncertainties, allowing for the decision to skip simulations where predictions are likely accurate, thereby reducing computational effort without compromising optimization accuracy. This module can implement Bayesian neural networks or Gaussian process models, refining its predictions with accurate simulation results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If surrogate models based on machine learning are used to predict simulation outcomes, then computational resources are reduced, but prediction accuracy deteriorates
Solution Approach 1:
A Bayesian machine learning model serves as an intermediary between design data and simulation results, providing predictions with associated uncertainty estimates. This intermediary allows the system to make informed decisions about which simulations to run, reducing overall computational resources while maintaining accuracy where predictions are reliable
Solution Approach 2:
The system dynamically changes the parameter of simulation execution by using uncertainty estimates to decide whether to run a full simulation or rely on the surrogate model prediction. When uncertainty is low, the surrogate model is used; when uncertainty is high, a simulation is performed, thus adapting computational effort to actual need
2Measurement precision
If extensive simulations are run for each design variant to ensure optimization accuracy, then optimization accuracy is maintained, but computational effort increases
Solution Approach 1:
Instead of running full simulations for all design variants, the system applies partial action by running simulations only for variants where the Bayesian model's uncertainty is high. For variants with low uncertainty, the surrogate model prediction is sufficient, thus avoiding excessive computational effort while maintaining optimization accuracy
Solution Approach 2:
The system uses feedback from the Bayesian model's uncertainty estimates to dynamically determine which design variants require full simulation. This feedback mechanism ensures that simulations are targeted at cases where they are most needed, maintaining optimization accuracy while reducing overall computational effort
Data Source
Figure 1~2
Figure 3~4
AI summary
A machine learning module (BNN) is provided which is trained to generate from a design data record (DR) specifying a design variant of a product (P), a first performance signal (PF1) quantifying a predictive performance of the design variant and a predictive uncertainty (UC) of the predictive performance. Furthermore, a variety of design data records (DR) each specifying a design variant of the product (P) is generated. For a respective design data record (DR), the following steps are performed: - a first performance signal (PF1) and a corresponding predictive uncertainty (UC) are generated, - depending on the predictive uncertainty (UC), a simulation (SIM) yielding a second performance signal (PF2) quantifying a simulated performance of the corresponding design variant is either run or skipped, and - a performance value (PV) is derived from the second performance signal (PF2) if the simulation (SIM) is run or, otherwise, from the first performance signal (PF1). Depending on the derived performance values (PV), a performance-optimizing design data record (ODR) is determined and output to control the production plant (PP).