Hybrid OLS for uncertainties estimation in direct shear testing. (November 2021)

• Record Type: Journal Article
• Authors: Ungureanu, C.; Priceputu, A.; Nguyen, G.; Pencea, I.; Turcu, R.N.; Popescu-Argeş, A.C.
• Is Part Of: Measurement. Volume 185(2021)
• Publication Date: 2021-11
• Subjects: Direct shear test -- Measurement uncertainty -- Hybrid ordinary least-squares approach -- Error-in-both-variables -- Error correlation -- Heteroscedasticity; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2021.110018
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: Uncertainty estimation is of great demand for geotechnics, but DST standards ignore it. OLS, IWLS and WLOC approaches are not suitable for uncertainty estimation in DST. Hybrid OLS (HOLS) approach is highly recommended for uncertainty estimation in DST. HOLS addresses correlated errors in both variables, including the heteroscedastic ones. Determination of MU in DST has a critical role in evaluation of soil improvement rate. Abstract: A new Hybrid Ordinary Least Squares method, denoted HOLS, is proposed for the measurement uncertainty estimation in direct shear tests performed in geotechnical laboratories. The HOLS is designed to overcome the shortcoming of the test standards that ignore the measurement uncertainty estimation in the geotechnical laboratory tests. HOLS is in line with the geotechnical laboratory practice, mathematically accessible and easy to deploy its related calculations on common PC software. The HOLS takes into account the correlations among all the measurands used in direct shear test procedure. It provides better results in both cases of the homoscedastic and heteroscedastic data compared to other approaches, as the Ordinary Least Squares largely overestimates measurement uncertainties, while Iterative Weighted Least Squares and Weighted Line of Organic Correlation strongly underestimate them. The proposed method was validated by inter-comparison between published data. The worst-case strategy is advanced in case of feeble knowledge about errors correlations.