فهرست نشریات علمی دسترسی آزاد ایران

Petroleum hydrocarbon contamination severely degrades soil engineering properties, including shear strength, compaction behavior, and hydraulic conductivity, thereby limiting its potential for sustainable reuse. This study experimentally evaluated the efficacy of microbial bioremediation on soils contaminated with 0–10% crude oil and developed predictive frameworks using Artificial Neural Networks (ANN) and Response Surface Methodology (RSM). Experimental results after 12 weeks of treatment revealed significant geotechnical recovery: Maximum Dry Density (MDD) increased by up to 5.1%, Optimum Moisture Content (OMC) decreased by 9.4%, and California Bearing Ratio (CBR) improved by over 137% compared to contaminated states. Specifically, remediated samples (e.g., BIO-03-12 and BIO-05-12) achieved CBR values > 8% and MDD >1,720kg/m3, satisfying standard subgrade requirements for low-to-medium traffic pavements. To address the limitations of the initial experimental dataset (n=17), data augmentation via Gaussian noise injection (σ = 0.03) was employed to expand the record to n=102 for high-fidelity modeling. Comparative performance analysis using Taylor diagrams demonstrated that while data augmentation was essential for the ANN, elevating its testing R2 from 0.8727 to 0.9856, the RSM exhibited superior inherent robustness on the non-augmented dataset (R2 = 0.9707). Furthermore, the RSM achieved a significantly lower Akaike Information Criterion (AIC) of 70.98 compared to the ANN’s 113.47, identifying it as the more parsimonious and reliable architecture for limited experimental data. These findings suggest that while augmented ANN models offer excellent forecasting for optimized treatment, the RSM remains the most defensible tool for direct application in small-scale geotechnical remediation projects.