Production Plant Control Using Pareto-Ranked Simulation Screening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing design optimization systems for complex products consume significant computational resources due to high simulation demands and often rely on inaccurate surrogate models, failing to efficiently optimize performance across multiple objectives.
Innovation Solution
Implement a machine learning module, such as a Bayesian neural network, to generate predictive performance values and uncertainties, allowing for the determination of Pareto fronts and selective simulation of design variants based on ranks and uncertainties, thereby reducing unnecessary simulations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If simulations are run for all design variants to ensure accurate performance evaluation, then measurement precision is improved, but computing time and computational resources increase significantly
Solution Approach 1:
The machine learning model is trained in advance on a subset of design variants to learn the relationship between design parameters and performance outcomes. This preliminary training enables the model to quickly predict performance for new design variants without requiring full simulations, thus reducing computing time while maintaining evaluation accuracy.
Solution Approach 2:
Instead of running expensive simulations for every design variant, the patent uses a machine learning model that creates a simplified copy or approximation of the simulation process. The model learns from simulated data and then predicts performance for new variants, replacing the need for repeated full simulations while preserving measurement precision.
2Loss of time
If surrogate models are used to reduce computational resources, then computing time is reduced, but measurement precision deteriorates due to poor accuracy
Solution Approach 1:
The machine learning model undergoes a preliminary training phase where it learns from a comprehensive set of simulated design variants. This training process enables the model to capture complex relationships and patterns in the data, improving its prediction accuracy for subsequent use as a surrogate model.
Solution Approach 2:
The patent employs advanced machine learning techniques that adjust model parameters and architecture to optimize prediction accuracy. By changing parameters such as model complexity, training data selection, and prediction methods, the system achieves high accuracy surrogate modeling that overcomes the limitations of traditional simplified models.
3Manufacturing precision
If all design variants are evaluated through simulation to identify optimal designs, then manufacturing precision is improved, but productivity decreases due to the large number of simulations required
Solution Approach 1:
The patent extracts only the most promising design variants for full simulation evaluation by using the machine learning model to pre-screen the design space. This extraction approach identifies a small subset of high-potential variants that warrant detailed simulation, maintaining design optimization accuracy while dramatically increasing productivity by avoiding unnecessary simulations of inferior designs.
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
An appropriately trained machine learning module (BNN) is used for generating a group of predictive performance values (PV) for each design data record (DR) of a first set (S1) of design data records. For the first set (S1), a pareto front (PF1) is determined based on the corresponding groups. In a further step, a first rank is assigned to the design data records (DR) of the pareto front (PF1). Then the design data records of the pareto front (PF1) are removed from the first set (S1), resulting in a reduced set (S2) of design data records. The above steps of determining a pareto front, assigning a rank, and removing design data records of a respective pareto front are repeated with decreasing ranks. Depending on a resulting rank of a respective design data record (DR), a simulation of the design variant corresponding to that design data record (DR) is either run or skipped. Based on an aggregated performance value (APV) derived from the simulation or from the corresponding group of predictive performance values (PV), a performance-optimizing design data record (ODR) is selected from the first set (S1) of design data records. The selected design data record (ODR) is then output for controlling the production plant (PP).