Scholarlink

<p>This paper introduces an entropy-assisted genetic algorithm (EGA) for global optimization and applies it to enhance Basis Gaussian Process Regression (BGPR) for fatigue prognosis in elastomeric rubber bearings, effectively addressing challenges posed by incomplete run-to-failure data. EGA optimizes the hyperparameters of BGPR by leveraging Renyi’s entropy to enhance population diversity, mitigating premature convergence, and ensuring global optimization. The framework enhances predictive accuracy by integrating BGPR with an adaptive basis function that dynamically updates based on detected degradation trends using Bayesian change point identification. While this approach enhances the BGPR framework’s adaptability, its reliance on predefined basis functions remains a constraint in achieving optimal predictive performance. The proposed approach is validated through numerical case studies and experimental fatigue data, demonstrating superior performance over conventional models. Results indicate an average improvement of 5% in predictive accuracy, particularly for non-stationary degradation patterns. The findings contribute to the advancement of data-driven structural health monitoring strategies, particularly for seismic isolation systems, enhancing the reliability and longevity of critical civil infrastructure subjected to fatigue-induced deterioration.</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

Prognosis of multi-stage degradation of elastomeric rubber bearing through entropy-assisted genetic optimization of Gaussian process regression

  • 期刊论文

  • 发表日期：

    2025年9月9日

查看全文
  • Shivam Ojha,
  • Amit Shelke

摘要

This paper introduces an entropy-assisted genetic algorithm (EGA) for global optimization and applies it to enhance Basis Gaussian Process Regression (BGPR) for fatigue prognosis in elastomeric rubber bearings, effectively addressing challenges posed by incomplete run-to-failure data. EGA optimizes the hyperparameters of BGPR by leveraging Renyi’s entropy to enhance population diversity, mitigating premature convergence, and ensuring global optimization. The framework enhances predictive accuracy by integrating BGPR with an adaptive basis function that dynamically updates based on detected degradation trends using Bayesian change point identification. While this approach enhances the BGPR framework’s adaptability, its reliance on predefined basis functions remains a constraint in achieving optimal predictive performance. The proposed approach is validated through numerical case studies and experimental fatigue data, demonstrating superior performance over conventional models. Results indicate an average improvement of 5% in predictive accuracy, particularly for non-stationary degradation patterns. The findings contribute to the advancement of data-driven structural health monitoring strategies, particularly for seismic isolation systems, enhancing the reliability and longevity of critical civil infrastructure subjected to fatigue-induced deterioration.