Scholarlink

<p>Multi-objective optimization problem (MOOP) presents inherent complexity compared to single-objective formulations due to conflicting objective trade-offs and substantial computational demands. To address these challenges, this study proposes an innovative six sigma framework integrating three key components: (1) density-based spatial clustering of applications with noise (DBSCAN) for solution space refinement, (2) a pseudo-single-loop strategy for iterative optimization, and (3) an enhanced random forest-deep neural network (RF-DNN) surrogate model. The methodology is validated through a lightweight passenger car seat design case study, where data scarcity is mitigated through RF-DNN-based augmentation, enabling high-precision surrogate modeling with limited initial samples. By synergistically combining DBSCAN clustering with the pseudo-single-loop strategy, computational efficiency is enhanced through dynamic elimination of non-dominant solutions during quality assessment phases. Experimental results demonstrate two key advancements: First, the proposed framework reduces computational time by 48.5% compared to conventional design for six sigma (DFSS) approaches while maintaining Pareto front quality. Second, the RF-DNN surrogate model achieves a minimum 10% improvement in prediction accuracy over standalone DNN and radial basis function (RBF) models. These findings highlight the framework’s potential to streamline complex engineering optimizations, particularly in resource-constrained industrial applications requiring rigorous multi-objective analysis.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Six Sigma Multi-objective Optimization Method Based on DBSCAN Clustering Pseudo-single-loop Strategy and Bayesian-Optimized RF-DNN

  • 期刊论文

  • 发表日期：

    2025年5月20日

查看全文
  • Huijie Yu,
  • Dasheng Zhao,
  • Xiangtian Wan

摘要

Multi-objective optimization problem (MOOP) presents inherent complexity compared to single-objective formulations due to conflicting objective trade-offs and substantial computational demands. To address these challenges, this study proposes an innovative six sigma framework integrating three key components: (1) density-based spatial clustering of applications with noise (DBSCAN) for solution space refinement, (2) a pseudo-single-loop strategy for iterative optimization, and (3) an enhanced random forest-deep neural network (RF-DNN) surrogate model. The methodology is validated through a lightweight passenger car seat design case study, where data scarcity is mitigated through RF-DNN-based augmentation, enabling high-precision surrogate modeling with limited initial samples. By synergistically combining DBSCAN clustering with the pseudo-single-loop strategy, computational efficiency is enhanced through dynamic elimination of non-dominant solutions during quality assessment phases. Experimental results demonstrate two key advancements: First, the proposed framework reduces computational time by 48.5% compared to conventional design for six sigma (DFSS) approaches while maintaining Pareto front quality. Second, the RF-DNN surrogate model achieves a minimum 10% improvement in prediction accuracy over standalone DNN and radial basis function (RBF) models. These findings highlight the framework’s potential to streamline complex engineering optimizations, particularly in resource-constrained industrial applications requiring rigorous multi-objective analysis.