Production Control Using Surrogate Ranking for Design Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer-aided design systems face inefficiencies in optimizing complex products due to high computational demands and inaccurate surrogate models, particularly in multi-dimensional optimization (MDO) methods, which require significant resources and often yield fluctuating accuracy.
Innovation Solution
A method involving the use of design evaluation modules, including machine learning modules like Bayesian neural networks, to predict target values with statistical distributions, combined with Pareto optimization to determine rankings, and a selection process to identify a target-optimized design data set, reducing computational effort while maintaining accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multidimensional optimization methods (MDO) are used to optimize target parameters, then optimization accuracy is improved, but computational effort increases considerably
Solution Approach 1:
The patent applies preliminary action by training surrogate models in advance to capture the relationship between design parameters and target parameters. These pre-trained models can then quickly predict optimization results without requiring repeated full simulations during the optimization process, thereby reducing computational effort while maintaining accuracy.
Solution Approach 2:
The patent creates simplified copies (surrogate models) of the complex simulation models. These surrogate models replicate the essential behavior of the original models but require significantly less computational resources, enabling efficient optimization while preserving optimization accuracy.
2Use of energy by stationary object
If surrogate models are used to reduce computational effort, then computational resources are reduced, but accuracy becomes low or fluctuating
Solution Approach 1:
The patent merges multiple surrogate models into an ensemble system. By combining the predictions of multiple models, the system achieves more stable and accurate results than individual models, reducing the fluctuating accuracy problem while maintaining low computational resource requirements.
Solution Approach 2:
The patent dynamically adjusts parameters of surrogate models based on the optimization progress and uncertainty estimates. This allows the models to adapt their complexity and accuracy levels, maintaining high prediction accuracy when needed while reducing computational resources during stable phases.
3Measurement precision
If a large number of design variants are evaluated to optimize target parameters, then optimization quality is improved, but evaluation time increases
Solution Approach 1:
The patent replaces the mechanical simulation evaluation process with surrogate model predictions. This substitution allows for rapid evaluation of design variants without executing time-consuming simulations, enabling the assessment of many more design variants within the same evaluation time while improving optimization quality.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
A plurality of test data sets is read in, which each comprise: - a first design data set (DR1), which specifies a design variant of a product (P); and - first target values (V1), which quantify target variables of said design variant which are to be optimized. With respect to the first target values (V1), a first ranking (R1) of the first design data sets (DR1) is determined. Furthermore, a plurality of design evaluation modules (EV1,...,EVN) for predicting target values on the basis of design data sets is provided. By means of the design evaluation modules (EV1,...,EVN), respective target values (VP1,...,VPN) for the first design data sets (DR1) are predicted. For each of the design evaluation modules (EV1,...,EVN), a second ranking (R2) of the first design data sets (DR1) with respect to the predicted target values (VP1,...,VPN) and a deviation (D) of the second ranking (R2) from the first ranking (R1) are then determined determined. One design evaluation module (EVS) is then selected in accordance with the determined deviations (D). Furthermore, a plurality of second design data sets (DR2) is generated, for each of which second target values (V2) are predicted by means of the selected design evaluation module (EVS). In accordance with the second target values (V2), a target-value-optimized design data set (ODR) is then derived from the second design data sets and is output for the manufacturing of the product (P).