Scholarlink

<p>Aiming at the problems of spectrum aliasing between fibers and asphalt matrix, difficult separation of multi-scale features in the spatial distribution detection of fiber asphalt mixture, as well as the defects of existing methods including high cost of physical scanning, large data dependence of deep learning and manual parameter trial of traditional signal processing. This paper takes polyvinyl alcohol (PVA) fiber asphalt pavement core samples from a highway as research objects, and proposes a fiber feature extraction method combining improved Enhanced Northern Goshawk Optimization (ENGO) and Warped Variational Mode Decomposition (WVMD). Multi-scale and multi-directional image decomposition is realized via Non-subsampled Shearlet Transform (NSST), and medium–high frequency sub-bands are screened by energy-entropy dual indicators and optimally converted into one-dimensional time-like signals. ENGO is improved by introducing cosine adaptive inertia weight, Cauchy mutation and collaborative fitness function of energy entropy-envelope entropy, achieving adaptive optimization of mode number (<i>K</i>) and penalty factor (<i>α</i>) of WVMD. Signal decomposition and reconstruction are completed by WVMD, and a fiber dispersion evaluation system is constructed combining amplitude and quantity features. Results show that the relative error of fiber bundle counting is 4.44% ~ 8.83%, and the dispersion scores of four core samples range from 66.66 to 98.26 with a relative error of 0.78% ~ 8.57% compared with measured data. The study&#xa0;reveals&#xa0;that better fiber dispersion in asphalt mixtures is associated with a smaller amplitude standard deviation, a&#xa0;narrower difference&#xa0;between the minimum and maximum amplitude, and a more reasonable statistical number of bundles. This method enables fast and accurate on-site fiber detection.</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

Spatial Distribution Detection of Fiber-Reinforced Asphalt Mixtures: Research on Multi-algorithm Fusion Extraction of Feature Signals

  • 期刊论文

  • 发表日期：

    2026年5月18日

查看全文
  • Xunqian Xu,
  • Cheng Zhou,
  • Shuyong Pan,
  • Lin Cheng,
  • Chen Zhang,
  • Xu Wu

摘要

Aiming at the problems of spectrum aliasing between fibers and asphalt matrix, difficult separation of multi-scale features in the spatial distribution detection of fiber asphalt mixture, as well as the defects of existing methods including high cost of physical scanning, large data dependence of deep learning and manual parameter trial of traditional signal processing. This paper takes polyvinyl alcohol (PVA) fiber asphalt pavement core samples from a highway as research objects, and proposes a fiber feature extraction method combining improved Enhanced Northern Goshawk Optimization (ENGO) and Warped Variational Mode Decomposition (WVMD). Multi-scale and multi-directional image decomposition is realized via Non-subsampled Shearlet Transform (NSST), and medium–high frequency sub-bands are screened by energy-entropy dual indicators and optimally converted into one-dimensional time-like signals. ENGO is improved by introducing cosine adaptive inertia weight, Cauchy mutation and collaborative fitness function of energy entropy-envelope entropy, achieving adaptive optimization of mode number (K) and penalty factor (α) of WVMD. Signal decomposition and reconstruction are completed by WVMD, and a fiber dispersion evaluation system is constructed combining amplitude and quantity features. Results show that the relative error of fiber bundle counting is 4.44% ~ 8.83%, and the dispersion scores of four core samples range from 66.66 to 98.26 with a relative error of 0.78% ~ 8.57% compared with measured data. The study reveals that better fiber dispersion in asphalt mixtures is associated with a smaller amplitude standard deviation, a narrower difference between the minimum and maximum amplitude, and a more reasonable statistical number of bundles. This method enables fast and accurate on-site fiber detection.