Scholarlink

For aerospace composite manufacturing and assembly, high confidence-level models are desired to predict and mitigate defects such as process-induced-deformations (PIDs). These are typically predicted with deterministic high-fidelity Finite Element (FE) process simulations, calibrated, and validated by conducting numerous experiments at different scales. However, this approach is very time-consuming, cost-intensive, and neglects the variabilities in the material and process, and their effects on manufacturing defects. To quantify such uncertainties, we propose a novel framework combining fast stochastic FE simulations with probabilistic machine learning. Results from fast FE simulations are used for Neural Network (NN) surrogate modeling, Global Sensitivity Analysis (GSA), and Markov Chain Monte Carlo (MCMC) to effectively quantify and visualize the effect of input uncertainties on the formation of manufacturing defects. This paper will demonstrate the capabilities of the proposed framework with a case study concerning material variabilities and manufacturing parameters such as cure cycle and tool geometry in the autoclave. Results contribute to better understanding of the propagation of uncertainties in the manufacturing process, therefore facilitating the prediction and mitigation of manufacturing defects.

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

Uncertainty Quantification in Advanced Aerospace Composite Manufacturing Through Stochastic Finite Element Analysis and Probabilistic Machine Learning

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Huilong Fu,
  • Kendall A. Johnson,
  • Navid Zobeiry

摘要

For aerospace composite manufacturing and assembly, high confidence-level models are desired to predict and mitigate defects such as process-induced-deformations (PIDs). These are typically predicted with deterministic high-fidelity Finite Element (FE) process simulations, calibrated, and validated by conducting numerous experiments at different scales. However, this approach is very time-consuming, cost-intensive, and neglects the variabilities in the material and process, and their effects on manufacturing defects. To quantify such uncertainties, we propose a novel framework combining fast stochastic FE simulations with probabilistic machine learning. Results from fast FE simulations are used for Neural Network (NN) surrogate modeling, Global Sensitivity Analysis (GSA), and Markov Chain Monte Carlo (MCMC) to effectively quantify and visualize the effect of input uncertainties on the formation of manufacturing defects. This paper will demonstrate the capabilities of the proposed framework with a case study concerning material variabilities and manufacturing parameters such as cure cycle and tool geometry in the autoclave. Results contribute to better understanding of the propagation of uncertainties in the manufacturing process, therefore facilitating the prediction and mitigation of manufacturing defects.