Inverse Design Optimization for Constrained Multi-Objective Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing system design methods are sample-inefficient and time-consuming, particularly when dealing with constrained multi-objective optimization problems, and they struggle to focus on specific regions near non-dominated solutions that are of most interest to designers.
Innovation Solution
A computer-implemented method for system optimization that uses a two-phase approach, involving a genetic algorithm with inverse design-based active learning to efficiently explore the design space and improve specific objectives and constraints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If evolutionary algorithms are used to solve CMOO problems, then diverse candidate solutions can be evaluated over multiple generations to avoid local minima, but the method is sample-inefficient and requires numerous runs to find the best solution
Solution Approach 1:
The patent introduces a surrogate model as an intermediary between the evolutionary algorithm and the expensive simulation. The surrogate model approximates the simulation responses, allowing the evolutionary algorithm to evaluate many candidate solutions without requiring numerous actual simulation runs. This mediator enables the algorithm to maintain its ability to explore diverse solutions while dramatically improving sample efficiency.
Solution Approach 2:
The patent performs preliminary action by training the surrogate model on a limited set of simulation data before running the evolutionary algorithm. This preliminary training phase prepares the surrogate model to provide accurate predictions during the optimization process, reducing the need for numerous simulation runs during the main optimization phase.
2Productivity
If Bayesian optimization is used for system design, then sample efficiency is improved through active learning, but the surrogate model building complexity becomes O(N3) where N is the number of samples
Solution Approach 1:
The patent replaces the complex Bayesian optimization surrogate model building process with a simpler machine learning-based surrogate model. Instead of using traditional BO methods with O(N3) complexity, the patent employs neural networks or other ML models that scale more favorably with the number of samples, substituting the mechanical BO process with a more efficient computational approach.
Solution Approach 2:
The patent changes the parameters of the surrogate model from traditional BO approaches to modern ML-based approaches. By using neural networks with appropriate architecture and training methods, the patent achieves comparable or better sample efficiency while reducing the computational complexity of model building from O(N3) to a more scalable form.
3Adaptability or versatility
If existing methods try to find the entire Pareto front, then comprehensive optimization coverage is achieved, but the method cannot focus on specific regions near non-dominated solutions that are of most interest to the designer
Solution Approach 1:
The patent applies local quality by allowing different regions of the design space to be optimized with different priorities. The surrogate model and evolutionary algorithm can focus computational resources on specific regions of interest near non-dominated solutions, while still maintaining awareness of the overall Pareto front. This enables targeted optimization in areas most important to the designer without completely abandoning comprehensive coverage.
Solution Approach 2:
The patent introduces dynamics by making the optimization process adaptive and flexible. The surrogate model is continuously updated during the evolutionary process, and the algorithm can dynamically shift focus to different regions of the design space based on designer preferences or emerging insights. This dynamic approach allows the method to balance comprehensive coverage with targeted optimization throughout the optimization process.
Data Source
AI summary
A design methodology and tool called INFORM are provided that use a two-phase approach for sample-efficient constrained multi-objective optimization of real-world nonlinear systems. In the first optional phase, one may modify a genetic algorithm (GA) to make the design process sample-efficient, and may inject candidate solutions into the GA population using inverse design methods. The inverse design techniques may be based on (i) a neural network verifier, (ii) a neural network, and (iii) a Gaussian mixture model. The candidate solutions for the next generation are thus a mix of those generated using crossover/mutation and solutions generated using inverse design. At the end of the first phase, one obtains a set of nondominated solutions. In the second phase, one chooses one or more solution(s) from the non-dominated solutions or another reference solution to further improve the objective function values using inverse design methods.