Bearing Capacity of Eccentrically Loaded Footings on Rock Masses Using Soft Computing Techniques

A crucial characteristic of real-world engineering operations is a strip footing's bearing capacity on a rock mass subjected to incline and eccentric loading conditions. Many scientists have attempted to establish and implement artificial intelligence (AI) models for estimating strip footings’ bearing capacity. In this study, four data-driven models, namely, extreme gradient boosting (XGBoost), random forest (RF), deep neural network (DNN), and long short-term memory (LSTM), are developed and compared to calculate the strip footing's bearing capacity. The strip footing's bearing capacity is obtained numerically by performing a lower bound (LB) and upper bound (UB) finite element limit analysis (FELA) for the purpose of training machine learning models. A total of 5120 FELA solutions with six dimensionless input parameters, namely, the geological strength index ( GSI ), the yield parameter ( m i ), the dimensionless strength ( 𝛾 B/ 𝜎 ci ) , inclination angle ( 𝛽 ), the dimensionless eccentricity ( e/B ), and the adhesion factor ( a ), and one output parameter, the bearing capacity factor ( P/ 𝜎 ci B ), were utilized in the analysis. The results show that the efficiency of all the proposed models is sufficient for bearing capacity factor determination, with coefficient of determination ( R 2 ) values ranging from 0.87 to 0.997 in the training phase and 0.975 to 0.999 in the testing phase. The proposed XGBoost model outperforms other models, such as RF, DNN, and LSTM, and can be used accurately for estimating a strip footing's bearing capacity on rock mass subjected to incline and eccentric loading loads.