Adaptive Classification for Multi-Objective Design Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional engineering design optimization methods, such as CAE analysis, rely heavily on trial-and-error and require numerous expensive experiments to achieve Pareto optimal solutions in multi-objective design optimization, making it inefficient for selecting design alternatives.
Innovation Solution
The method employs adaptive classification using a multi-dimensional space division scheme, like support vector machines, to partition the design space and select non-dominated design alternatives, repeatedly evaluating and refining these selections until an end condition is reached, thereby efficiently identifying Pareto optimal solutions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional CAE analysis and trial-and-error methods are used for multi-objective design optimization, then design alternatives can be evaluated, but the number of expensive experiments required increases significantly
Solution Approach 1:
The design space is partitioned into multiple sub-regions using space division schemes (e.g., support vector machines). This segmentation allows the optimization process to focus on specific regions containing non-dominated solutions, reducing the overall number of experiments needed while maintaining accuracy in identifying the Pareto optimal front.
Solution Approach 2:
The method performs preliminary classification of the design space to identify regions likely to contain non-dominated solutions before conducting full evaluations. By pre-selecting promising regions and design alternatives, the approach avoids expensive experiments in regions unlikely to yield optimal solutions, thereby reducing computational time and cost.
2Adaptability or versatility
If the design space is thoroughly explored to ensure all Pareto optimal solutions are found, then solution completeness is improved, but the complexity of the optimization process increases
Solution Approach 1:
By dividing the design space into manageable sub-regions, the method maintains solution completeness while reducing process complexity. Each sub-region can be evaluated independently, allowing systematic exploration without overwhelming computational burden.
Solution Approach 2:
The approach introduces a new dimension of classification by evaluating design alternatives based on non-domination criteria within partitioned spaces. This additional dimensional perspective allows the method to identify Pareto optimal solutions more efficiently without requiring exhaustive search of the entire design space.
3Measurement precision
If more design alternatives are evaluated to improve the accuracy of Pareto optimal front identification, then solution precision is improved, but the computational cost increases
Solution Approach 1:
The method performs preliminary space partitioning and classification to identify regions containing non-dominated solutions before conducting detailed evaluations. This preliminary action ensures that computational resources are focused on evaluating design alternatives that are most likely to contribute to an accurate Pareto optimal front, rather than wasting resources on dominated alternatives.
Solution Approach 2:
The optimization process applies different evaluation strategies to different regions of the design space. Regions identified as containing non-dominated solutions receive more thorough evaluation, while other regions receive less intensive processing. This local differentiation maintains accuracy in identifying the Pareto optimal front while reducing overall computational cost.
Data Source
AI summary
Definition of a design space and an objective space for conducting multi-objective design optimization of a product is received in a computer system having a design optimization application module installed thereon. Design space is defined by design variables while objective space is defined by design objectives. First set of designs in the design space is selected. Each of the first set is evaluated in the objective space for non-dominance. Design space is partitioned into first and second regions using a multi-dimensional space division scheme (e.g., SVM). The first region is part of the design space containing all of the non-dominated design alternatives while the second region contains remaining of the design space. Second set of designs is selected within the first region. Each of the second set and existing non-dominated design alternatives are evaluated for non-dominance. Multi-objective optimization repeats the partition and evaluation until an end condition is reached.