A Simulation Approach to Predict Uniaxial Compressive Strength of Shale and Sandstone Samples Using Artificial Neural Network

Proper determination of Unconfined Compressive Strength (UCS) of rocks is a crucial subject in design of geotechnical structures. Although direct determination of UCS through laboratory test appears to be relatively simple, obtaining proper core segments specifically for weathered rocks is difficult and expensive. It is well established that UCS can be estimated indirectly using rock index properties. In comparison to the direct test, indirect prediction of UCS is relatively easier and cheaper. This study involves extensive laboratory tests on 32 datasets of shale and sandstone in various weathering grades obtained from excavation site in Johor, Malaysia. The laboratory tests include UCS test, Brazilian Tensile Strength (BTS) test, Point Load Index Test (Is(50)), P-wave velocity (Vp) test Schmidt Hammer Rebound Number (Rn) and Dry Density (DD) measurement. The application of Artificial Neural Network (ANN) in UCS prediction is highlighted in this study. For this reason, BTS, Is (50), Vp, Rn and DD were considered as input parameters while the UCS was set to be the output. The ANN results shows the superiority of ANN in UCS prediction.

Unconfined Compressive Strength, Laboratory Tests, Artificial Neural Network.