Scholarlink

<p>This study proposes an integrated, uncertainty-aware framework for estimating pipeline flowrate from radiotracer residence-time-distribution (RTD) signals and evaluates it on industrial data. The novel contributions are threefold: a unified workflow that begins with digital signal processing, extracts RTD features, and then applies machine-learning (ML) prediction while reporting both point estimates and quantified uncertainty (bootstrap and Bayesian credible intervals); a systematic mean-residence-time (MRT)-centric analysis showing greater stability than peak-time methods across detector pairs; and a comparative assessment of two field data-acquisition systems (Ludlum and Ashtar/ALTIX) using Tc-99&#xa0;m injections (5–10&#xa0;mCi) recorded by scintillation detectors. RTD features (MRT, FWHM, peak amplitude, spectral descriptors) are mapped to flowrate using support-vector regression, Random Forest, and gradient-boosted trees. The machine-learning layer captures nonlinear relations between RTD features and flowrate that are not recovered by peak-based calculations, while the uncertainty layer assigns calibrated confidence to each estimate. Overall, an MRT-centric, ensemble-learning approach improves the reliability and interpretability of RTD-based flow estimation and is suitable for on-line investigations using radiotracer technology.</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

Efficient processing algorithms of radiotracer signals for optimizing flowrate measurements based on residence time distribution

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Elsayed H. Ali,
  • Horeya A. Arafa,
  • H. Kasban,
  • Mohamed S. El Tokhy

摘要

This study proposes an integrated, uncertainty-aware framework for estimating pipeline flowrate from radiotracer residence-time-distribution (RTD) signals and evaluates it on industrial data. The novel contributions are threefold: a unified workflow that begins with digital signal processing, extracts RTD features, and then applies machine-learning (ML) prediction while reporting both point estimates and quantified uncertainty (bootstrap and Bayesian credible intervals); a systematic mean-residence-time (MRT)-centric analysis showing greater stability than peak-time methods across detector pairs; and a comparative assessment of two field data-acquisition systems (Ludlum and Ashtar/ALTIX) using Tc-99 m injections (5–10 mCi) recorded by scintillation detectors. RTD features (MRT, FWHM, peak amplitude, spectral descriptors) are mapped to flowrate using support-vector regression, Random Forest, and gradient-boosted trees. The machine-learning layer captures nonlinear relations between RTD features and flowrate that are not recovered by peak-based calculations, while the uncertainty layer assigns calibrated confidence to each estimate. Overall, an MRT-centric, ensemble-learning approach improves the reliability and interpretability of RTD-based flow estimation and is suitable for on-line investigations using radiotracer technology.