Field Study on Unconfined Compressive Strength and Drilling Data of DSM Columns: A Machine Learning Approach

Numerous limitations of proper lands, especially in offshores, bring about outcomes to consider soil improvement methods. The deep soil mixing (DSM) method is an efficient ground improvement procedure using in situ soils for treatment plans. In order to ensure the quality of the treated soil, the DSM elements should be evaluated using unconfined compressive strength (UCS) tests on only 2%-4% of executed columns, based on the recommendations of the FHWA. Therefore, there are significant uncertainties regarding the UCS of the other 96% of the untested DSM elements. To reduce these deficiencies and bring an insight into the uncored DSM columns indirectly, artificial intelligence (AI) and machine learning (ML) methods (XGBoost, random forest, artificial neural network, and lasso regression) were implemented to predict a correlation algorithm between UCS results and several drilling rig machine parameters, including drilling, lifting, rotary speeds, rotary pressures, and the added water during pre-drilling phase. The extensive results of over 3,780 UCS tests and corresponding machine data of a DSM soil improvement project in Hormozgan, Iran, were considered to train and test the ML algorithms. In this case, instead of conducting a probability analysis to find out a DSM column UCS for using in design procedure, the predicted strength of each column by the well-trained ML method can be considered.