Production Control Using Cross-Version ML Design Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current multidimensional optimization methods for complex product design are computationally intensive due to the need to evaluate numerous design variants, and surrogate models often exhibit low or variable accuracy, making it inefficient to optimize product performance.
Innovation Solution
A method and system that utilize machine learning modules to jointly train on data from multiple product versions, allowing for the generation of synthetic design data sets and optimization of design data for a new product version by leveraging interdependencies between design and performance across versions, while accounting for version-specific differences through product version-specific modules.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multidimensional optimization methods simulate a large number of design variants to optimize product performance, then the accuracy of performance optimization is improved, but the computational effort and time required increase significantly
Solution Approach 1:
The system performs preliminary simulation of design variants to generate training data sets before the actual optimization process. These pre-simulated data sets are stored and reused to train surrogate models, eliminating the need to perform full simulations during the optimization phase. This preliminary action significantly reduces computational time while maintaining optimization accuracy.
Solution Approach 2:
The system creates surrogate models that are trained copies of the full simulation system. These surrogate models replicate the behavior of complex simulations but execute much faster. Once trained on simulation data, the surrogate models can predict performance metrics without requiring computationally intensive full simulations, thus reducing time loss while preserving accuracy.
2Loss of time
If surrogate models are used to predict simulation results without detailed simulation to reduce computational effort, then the computational time is reduced, but the accuracy of performance prediction decreases or becomes variable
Solution Approach 1:
The system merges multiple surrogate models into an ensemble that collectively provides more accurate predictions than individual models. By combining predictions from multiple trained surrogate models, the system reduces the variability and improves the overall accuracy of performance predictions while maintaining the computational efficiency of surrogate modeling.
Solution Approach 2:
The system implements feedback mechanisms where prediction results from surrogate models are validated against actual simulation outcomes or experimental data. This feedback is used to iteratively refine and retrain the surrogate models, improving their accuracy over time while maintaining the speed advantage of surrogate modeling.
3Measurement precision
If product version-specific machine learning modules are trained separately to account for version differences, then the prediction accuracy for each version is improved, but the training effort and data requirements increase
Solution Approach 1:
The system segments the training process into two distinct phases: first, a common machine learning model is trained on data from multiple product versions to learn universal patterns; second, version-specific adjustments are made by training separate modules on version-specific data. This segmentation allows the system to benefit from shared knowledge across versions while still capturing version-specific characteristics, improving efficiency compared to training completely separate models from scratch.
Solution Approach 2:
The system develops a universal machine learning model that can handle multiple product versions through transfer learning. The model trained on data from one product version can be adapted and fine-tuned for other versions, reducing the overall training effort. This universal model serves multiple functions across different product versions, improving productivity while maintaining version-specific accuracy through fine-tuning.
Data Source
AI summary
For controlling a production system product version-specific training data sets are read in for each of multiple product versions. Each training data set comprises a design data set. The design data sets are fed into a machine learning module covering all product versions. An output signal is fed into both a first product version-specific machine learning module and also a second product version-specific machine learning module. The machine learning modules are jointly trained so that output data (O1) of the first machine learning module reproduces the performance values of the first product version and output data of the second machine learning module reproduces the performance values of the second product version. Then, a plurality of synthetic design data sets are generated and fed into the trained machine learning module. The resulting output signal is fed into the trained first machine learning module. A performance-optimized design data set is derived.