| Abstrakt: |
This study developed artificial intelligence models to predict Atterberg's limits, specifically the liquid (LL) and plastic limits (PL), based on #200 sieve analysis, which is a laborious and challenging task in the laboratory. Conventional methods use #40 (0.425 mm) sieve material, which contains fine sand that causes discrepancies, whereas using #200 (0.075 mm) sieve material is essential for accurate LL/PL determination. This study introduces novel Artificial Neural Network (ANN) and Multivariate Regression (MLR) based models for LL/PL to mitigate these constraints. For this study, soil samples were collected from 120 locations, and tests such as sieve analysis, hydrometer analysis, and Atterberg’s limits using #40 and #200 sieves were conducted. Then, AI-based predictive models were developed. The results showed significantly higher LL/PL values for #200 sieve material than #40, with about 45% of samples experiencing changes in soil classification due to altered plasticity levels. The suggested forecasting LL model yielded R2and RMSE values of 0.999 and 1.162, respectively, while the PL data showed values of 0.999 and 0.974. These outcomes affirm the intended forecasting model's superior precision and generalization capacity. The prediction models have undergone stringent statistical validation and meet key performance metrics. Furthermore, Pearson correlation, sensitivity analysis, and parametric investigation highlight fine content, LL40, and PL40as critical determinants of ALs for sieve #200 passing materials. |
Full text from SpringerLink