The rock unconfined compressive strength (UCS) is considered one of the most important rock geomechanical properties as it is practically used in designing drilling programs and reservoir fracture jobs. The practical way for determining the UCS by lab measurements is costly and time-consuming, in addition, the empirical correlations require logs data for UCS estimation. The main objective of this study is to employ machine learning techniques for predicting the rock UCS from only the surface drilling parameters for complex lithology. The study provides UCS predictive model by using an artificial neural network (ANN). The data used was collected during drilling different formations with a complex lithology. A cleaned data set (2,926 measurements) was used for building the ANN model. The model was trained, tested, and optimized to provide high accuracy prediction for UCS. The results showed an overall strong UCS prediction with a correlation coefficient (R) greater than 0.99 and less than 5.52% an average absolute percentage error (AAPE). Furthermore, the model was validated with unseen data set and proved the high accuracy performance level (R of 0.99 and AAPE of 6.9%) that enhance the model application for UCS prediction in the practical drilling operations that will save extra cost and time.
Building the reservoir geomechanical model requires the rock strength UCS data that represents the maximum limit for the rock compressive strength without failure in case of applying uniaxial load under a confinement condition for the rock geomechanics testing (Chau et al., 1996). UCS is defined when no confinement is applied thus unconfined compressive strength. UCS is one of the rock failure parameters and this parameter should be determined with high accuracy to avoid most of the downhole drilling problems (Fjar et al., 2008). Also, based on the acquiring of UCS data, the drilling performance is optimized in terms of bit hydraulics, proper mud weight to mitigate the wellbore instability problems (Shi et al., 2015). Many geomechanics lab testing can be performed to determine the rock strength (UCS). Such lab testing required specific core sampling procedures to represent the real stress state condition of the formation rock. The lab measurement for the rock strength is considered the best precise value for the rock UCS and it is dependable in the program design for drilling and modeling (Liu, 2017). Determining UCS from the lab testing is costly, time-consuming, and usually performed to specific sections, and hence, not a complete UCS profile (Abdulraheem et al., 2009). Thus, many studied were accomplished to provide empirical relations for UCS with the rock petrophysical properties (Militzer and Stoll, 1973; Golubev and Rabinovich, 1976; Chang et al., 2006; Mostofi et al., 201; Nabaei and Shahbazi, 2012; Amani and Shahbazi, 2013a). These correlations mainly depend on the sonic transit time and the rock porosity for determining the rock UCS; however, these logs are not available in all wells or during the drilling operations.
