PREDICTION AND RELIABILITY ANALYSIS OF SOIL STRENGTH (CBR) USING MULTIPLE REGRESSION ANALYSIS

Abstract

The California Bearing Ratio (CBR) test is a penetration test used to evaluate the subgrade strength of roads and pavements. Results from the CBR test are used with empirical curves to determine the thickness of pavement layers, and the test remains the most widely used method for the design of flexible pavements. Under normal conditions, obtaining the result of a soaked CBR test takes between four and five days, a duration that can be impractical in emergency situations. During the COVID-19 pandemic, for example, pressure to deliver infrastructure quickly led to the construction of a 1,000-bed Huoshenshan Hospital in Wuhan, China, in ten days (24 January to 2 February 2020); achieving comparable feats depends on faster means of estimating soil strength. This study therefore explores an indirect, regression-based approach to predicting CBR. Geotechnical test results for 84 soil samples were collected, and a multiple regression analysis was carried out with the soaked CBR value as the dependent variable and existing moisture content (EMC), liquid limit (LL), plastic limit (PL), plasticity index (PI), particle size distribution, optimum moisture content (OMC), and maximum dry density (MDD) as independent variables. The analysis produced the relationship Y = −0.57359169X₁ + 0.677530201X₂ – 0.64691219X₃ – 0.59578021X₄ + 0.004879034X₅ – 0.01717437X₆ + 0.13710102X₇ – 0.266139X₈ – 0.10248341X₉ + 0.118277058X₁₀ – 0.10766766X₁₁ + 0.25140518X₁₂ – 0.16878082X₁₃ + 7.543637025X₁₄ – 0.20878597X₁₅ + 5.381442462X₁₆ + 10.18311796, where Y is the predicted soaked CBR value and X₁ through X₁₆ are the corresponding independent variables. The equation was used to predict CBR values for the remaining samples, achieving an accuracy of approximately 92%. A reliability analysis was subsequently carried out on each independent variable; the percentage passing sieve No. 200 and the maximum dry density returned R² values of 0.9872 and 0.9691, respectively, against the coefficient of variation.