Scholarlink

<p>Wire arc additive manufacturing (WAAM) is susceptible to arc instability, spatter, and porosity that deteriorate the quality. This method letter presents an acoustic-driven framework integrating a Gaussian Hidden Markov Model (HMM) with a contextual bandit reinforcement learning (RL) controller for layer-wise process monitoring and adaptive decision support. An 11- state HMM identifies latent deposition regimes; automatic Gaussian Mixture Model (GMM) thresholding classifies layer stability without manual tuning. An ε-greedy bandit learns state-conditioned corrective actions offline. The framework correctly detects a known material transition event (LA100S–ER2209, Layers 44–47) without prior knowledge of its location, corroborated by independent physical characterization [9], demonstrating practical utility for WAAM process monitoring.</p> Graphical abstract <p></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

Acoustic-driven Hidden Markov and reinforcement learning framework for intelligent process monitoring and adaptive decision support in wire arc additive manufacturing

  • 期刊论文

  • 发表日期：

    2026年6月2日

查看全文
  • Md Arifur Rahman,
  • Hossein Taheri

摘要

Wire arc additive manufacturing (WAAM) is susceptible to arc instability, spatter, and porosity that deteriorate the quality. This method letter presents an acoustic-driven framework integrating a Gaussian Hidden Markov Model (HMM) with a contextual bandit reinforcement learning (RL) controller for layer-wise process monitoring and adaptive decision support. An 11- state HMM identifies latent deposition regimes; automatic Gaussian Mixture Model (GMM) thresholding classifies layer stability without manual tuning. An ε-greedy bandit learns state-conditioned corrective actions offline. The framework correctly detects a known material transition event (LA100S–ER2209, Layers 44–47) without prior knowledge of its location, corroborated by independent physical characterization [9], demonstrating practical utility for WAAM process monitoring.

Graphical abstract