Evaluation of Uniaxial Compressive Strength from Physical Properties of Rocks to Obtain Formation Hardness And Bit Design Selection

Abstract

Uniaxial Compressive Strength (UCS) is the ultimate stress that a sample can stand before
breaking or failing.The traditional method of for finding Uniaxial Compressive Strength is known to be
costly, time consuming and destructive to the core samples being tested. In this project, a Neural Network
was produced to correlate between UCS and different rock physical properties for sandstone and limestone
samples from various data collected from previous research. Different models were created to be able to
investigate the best conditions under which the network is most optimized. The final network was created
using 18 hidden neurons with input parameters being porosity, p-wave velocity and rock type and the output
being UCS. The correlation coefficient from this network was found to be 0.975. The main factors found to
significantly affect the efficiency of the correlation or network were formation age, input parameters being
used, and the type of model used for statistical analysis. Results produced identified that narrowing the range
of difference in age between the samples being tested significantly improves the strength of the network
specifically from 0.870 to 0.968. Also, density was found to have the least correlation with UCS with a
correlation of coefficient of 0782 followed by p-wave velocity at 0.818 and the strongest correlation found
between UCS and proosity at a correlation coefficient of 0.862. Comparing Simple Regression model
(SRM), Multiple Regression Model (MRM) and Artificial Neural Network(ANN) the use of a Neural
Network was found to produce the best correlation coefficient and best RMSE values.