Scholarlink

<p>To assess the quality and initial damage of indoor sandstone samples, an integrated framework combining ultrasonic testing, mechanical testing, and machine learning was developed. Initially, waveforms and peak stresses of 207 samples were acquired. Subsequently, 29 features extracted from time-domain waveforms and the wavelet packet energy were utilized to construct a dataset. Feature weights determined by the Spearman rank correlation matrix were incorporated into preprocessing. Following this, two cluster analyses employing the Gaussian mixture model were conducted on the original and preprocessed datasets, with t-distributed stochastic neighbor embedding applied for visualization. Critically, the preprocessed dataset produced final clustering results showing denser intra-cluster scatter distributions and enhanced clustering metrics: mean silhouette coefficient (0.624), Calinski–Harabasz index (293.413), and Davies–Bouldin index (0.617). Consequently, the feature-weighted clustering outcome was adopted for quality classification. Furthermore, calculation methods for the quality score and relative damage level were proposed, establishing a classification scheme. Validation through multivariate analysis of variance (MANOVA) and Spearman rank correlation confirmed the scheme’s rationality and applicability. MANOVA indicated significant discrimination (<i>p</i> &lt; 0.001) in the combined means of peak stresses and P-wave velocities across quality grades. Spearman analysis demonstrated strong positive correlations (<i>p</i> &lt; 0.001) between quality grades and these core indicators. Concurrently, representative samples exhibited variations in key wavelet time–frequency characteristics (amplitude, the frequency corresponding to maximum amplitude, and the time corresponding to maximum amplitude) across ascending quality grades; these variations aligned with wave propagation behavior in rock media, further validating the scheme’s reliability.</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

Quality Classification of Indoor Rock Samples Using Ultrasonic Waveform Features and Cluster Analysis

  • 期刊论文

  • 发表日期：

    2025年11月11日

查看全文
  • Siqing Lv,
  • Jianfeng Liu,
  • Huining Xu,
  • Yang Wu,
  • Dehang Liu,
  • Junjie Liu

摘要

To assess the quality and initial damage of indoor sandstone samples, an integrated framework combining ultrasonic testing, mechanical testing, and machine learning was developed. Initially, waveforms and peak stresses of 207 samples were acquired. Subsequently, 29 features extracted from time-domain waveforms and the wavelet packet energy were utilized to construct a dataset. Feature weights determined by the Spearman rank correlation matrix were incorporated into preprocessing. Following this, two cluster analyses employing the Gaussian mixture model were conducted on the original and preprocessed datasets, with t-distributed stochastic neighbor embedding applied for visualization. Critically, the preprocessed dataset produced final clustering results showing denser intra-cluster scatter distributions and enhanced clustering metrics: mean silhouette coefficient (0.624), Calinski–Harabasz index (293.413), and Davies–Bouldin index (0.617). Consequently, the feature-weighted clustering outcome was adopted for quality classification. Furthermore, calculation methods for the quality score and relative damage level were proposed, establishing a classification scheme. Validation through multivariate analysis of variance (MANOVA) and Spearman rank correlation confirmed the scheme’s rationality and applicability. MANOVA indicated significant discrimination (p < 0.001) in the combined means of peak stresses and P-wave velocities across quality grades. Spearman analysis demonstrated strong positive correlations (p < 0.001) between quality grades and these core indicators. Concurrently, representative samples exhibited variations in key wavelet time–frequency characteristics (amplitude, the frequency corresponding to maximum amplitude, and the time corresponding to maximum amplitude) across ascending quality grades; these variations aligned with wave propagation behavior in rock media, further validating the scheme’s reliability.