Estudo da variabilidade estatística de ensaios de classificação de solos.

Abstract

The study presented in this dissertation deals with the reproducibility of four soil classification trials. A critical review of the literature, test methods and reproducibility results obtained in other countries is presented. The laboratory study was carried out using eleven natural soils, three industrially benefited soils and two soils resulting from natural and beneficiated soil mixtures. The reproducibility of the Liquidity Limit assay, Plasticity Limit assay, Granulometric Analysis and Real Density assay is presented in terms of repeatability and reproducibility. The repeatability study was conducted in a laboratory with the same equipment and the same operator, while the reproducibility study was supported by sixteen laboratories (three from universities, eight from federal and state road organizations, and five from consulting firms). A pilot study was also carried out to investigate the advantages of using the Penetration Cone method to determine the Liquidity Limit of fourteen soils of the sixteen investigated. Statistical analysis of the repeatability study shows that the coefficients of variation are comparable with other reported in the literature, while the reproducibility results show very clearly that the coefficients of variation are extremely unacceptable. It is shown that the reproducibility of the four soils surveyed is so poor that its validity is disputed. A discussion of the methods used is presented and suggestions are made to improve the reproducibility of the assays.