Multivariate Artificial Neural Network (ANN) models for predicting uniaxial compressive strength from index tests

Othman B. S., Ozcan N., KALENDER A., SÖNMEZ H.

International European Rock Mechanics Symposium (EUROCK), Saint Peter, Guernsey And Alderney, 22 - 26 May 2018, pp.345-351 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Volume:
  • City: Saint Peter
  • Country: Guernsey And Alderney
  • Page Numbers: pp.345-351
  • Hacettepe University Affiliated: Yes


Uniaxial Compressive Strength (UCS) of rock material is an important parameter used as input for rock engineering applications. However, preparation of standard test samples to determine UCS from some rock materials such as thinly bedded, weak and jointed rock masses are almost impossible. Therefore some index and indirect tests such as Block Punch Index (BPI), point load index (I-850) and Brazilian tensile strength have been proposed in order to estimate the UCS of such rocks. In the literature, the relations between UCS and index tests were extensively investigated by using statistical models. In fact, different rock materials with similar index test results may have different UCS's depending on their failure envelopes. By considering this fact, multivariate equation models were developed in this study to estimate UCS values. A large database was prepared with data compiled. In addition to developed simple empirical equations, Artificial Neural Network (ANN) method was used to develop models for prediction of UCS considering BPI, I-8(50), tensile strength (sigma(t)), gamma and mi parameters. Although the prediction models developed in this study may be considered for practical purposes, the results indicated that the models using BPI as an input parameter have higher prediction performance than others.