Formation density plays a central role to identify the types of downhole formations. It is measured in the field using density logging tool either via logging while drilling (LWD) or more commonly by wireline logging, after the formations have been drilled, because of operational limitations during the drilling process that prevent the immediate acquisition of formation density.
The objective of this study is to develop a predictive tool for estimating the formation bulk density (RHOB) while drilling using artificial neural networks (ANN). The ANN model uses the drilling mechanical parameters as inputs and petrophysical well-log data for RHOB as outputs. These drilling mechanical parameters including the rate of penetration (ROP), weight on bit (WOB), torque (T), standpipe pressure (SPP) and rotating speed (RPM), are measured in real time during drilling operation and significantly affected by the formation types. A dataset of 2,400 data points obtained from horizontal wells was used for training the ANN model. The obtained dataset has been divided into a 70:30 ratio for training and testing the model, respectively.
The results showed a high match with a correlation coefficient (R) between the predicted and the measured RHOB of 0.95 and an average absolute percentage error (AAPE) of 0.71%. These results demonstrated the ability of the developed ANN model to predict RHOB while drilling based on the drilling mechanical parameters using an accurate and low-cost tool. The black-box mode of the developed ANN model was converted into white-box mode by extracting a new ANN-based correlation to calculate RHOB directly without the need to run the ANN model. The new model can help geologists to identify the formations while drilling. Also, by tracking the RHOB trends obtained from the model it helps drilling engineers avoid many interrupting problems by detecting hazardous formations, such as overpressured zones, and identifying the well path, especially while drilling horizontal sections. In addition, the continuous profile of RHOB obtained from the developed ANN model can be used as a reference to solve the problem of missing and false logging data.