探討圓錐貫入試驗所得錐尖阻抗值之異常值

Abstract

本研究旨在針對圓錐貫入試驗（Cone Penetration Test, CPT）中所量測的錐尖阻抗力 (qc) 進行深入的異常值分析，以提升土壤力學性質評估的精確性和可靠性。CPT試驗作為一種常見的現地調查技術，能夠快速獲取地層的垂直剖面資料，有助於全面了解地層結構的變化。試驗過程中，圓錐形探頭以2 cm/sec的速率貫入土壤，並即時記錄錐尖阻抗力(qc) 和套筒側壁摩擦力(fs) 等關鍵參數。本研究主要聚焦於不同場址CPT資料中錐尖阻抗力(qc) 的異常值，因這些異常值可能會對後續的土壤特性分析及不確定性。 在研究中，本研究首先基於現地量測的錐尖阻抗力(qc) 建立隨機場模型，並採用過渡馬可夫鏈蒙地卡羅（TMCMC）演算法進行模擬，模擬出各項參數的合理範圍值，包括σ、𝜈z、𝜈ℎ、𝛿z、𝛿ℎ、𝜎𝑡、𝛿z𝑡、𝛿ℎ𝑡 等，藉此模擬錐尖阻抗力(qc)的合理範圍。研究進一步針對單孔及鄰近多孔的錐尖阻抗力(qc)資料進行異常值分析。在單孔分析中，先透過 WTMM 方法對土壤進行分層，而在鄰近多孔分析中，運用 k-means clustering 進行土壤趨勢分類，以降低不同土壤類型之間趨勢差異對異常值分析的影響。 研究結果表明，本研究提出的異常值分析方法在錐尖阻抗力(qc)的異常值檢測上符合視覺判斷之結果，證實其有效性。這種自動化分析方法不僅能大幅減少依賴人為判斷所需的時間，還能降低由於人為判斷不一致或疏失所造成的誤差。透過此方法，可以更加精確地檢測異常值，並提高數據處理的效率和可靠性，為後續的土壤特性分析提供更穩定的依據。

This study focuses on the in-depth analysis of outliers in the cone tip resistance (qc) measurements obtained from Cone Penetration Tests (CPT) to enhance the accuracy and reliability of soil mechanical property evaluations. As a widely used in-situ testing technique, CPT allows for the rapid acquisition of vertical soil profiles, helping to better understand stratigraphic variations. During the test, a cone-shaped probe is pushed into the soil at a rate of 2 cm/sec, continuously recording key parameters such as cone tip resistance (qc) and sleeve friction (fs). This study primarily investigates the outliers in qc measurements from different sites, as these outliers may introduce uncertainties in subsequent soil property analyses. In this research, a random field model was established based on the measured qc values, and the Transition Markov Chain Monte Carlo (TMCMC) algorithm was employed to simulate the reasonable range of parameters, including σ、𝜈z、𝜈ℎ、𝛿z、𝛿ℎ、𝜎𝑡、𝛿z𝑡、𝛿ℎ𝑡 ,to estimate the reasonable range of qc. Further outlier analysis was conducted on both single boreholes and neighboring boreholes. For single borehole analysis, the WTMM method was used for soil stratification, while for neighboring boreholes, k-means clustering was applied to classify soil trends, reducing the influence of variations between different soil types on outlier detection. The results demonstrate that the proposed outlier detection method is consistent with visual judgment, confirming its effectiveness. This automated method significantly reduces the time required for manual assessment and minimizes errors caused by inconsistent or subjective human judgments. It enables more precise detection of outliers and improves the efficiency and reliability of data processing, providing a more stable foundation for subsequent soil property analyses.