Showing papers by "Evren Ozbayoglu published in 2022"

Cement Sheath Fatigue Failure Prediction by Support Vector Machine Based Model

Abstract: Zonal isolation is significant for safety operation of the well. Failure to keep wellbore integrity can lead to sustained annulus pressure (SAP), and gas migration (GM), which may cause long non-productive time. Losing zonal isolation can cause severe environmental issue, which is irreversible and detrimental. However, cement sheath is exposed to temperature and pressure changes from the beginning of the drilling process to the whole life of the well. These cyclic changes can lead to fatigue failure of the cement. The objective of this study is to investigate the fatigue failure that caused by cyclic changing of temperature and pressure during life of the well. The scope of the study is based on the laboratory fatigue failure cases in previous literatures. Instead of using mechanical failure models, support vector machine (SVM) model is used to predict the fatigue failure of the cement sheath. The data is gathered from six papers of One-Petro, which includes 325 laboratory cement fatigue failure cases. The model has fourteen inputs. Seven cement related factors were selected, which include cement type, additive material, Uniaxial Confining Strength (UCS), curing temperature, curing pressure, curing age, and Young's modulus. Seven experimental related factors, which include highest inner pressure, loading increment rate, frequency of loading, experimental temperature, confining pressure, existence of outer confining part, and cycles to reach failure. The SVM model is implemented by Python. We investigated 240 combinations of input groups and selected the best performance SVM model. The classification result is zero for no fatigue failure, and one for failure. The accuracy for the SVM model is 72.7%, which shows that SVM can be an acceptable model for cement fatigue prediction. The SVM model we proposed is more applicable for real implementation. Because we used real wellbore geometry data (thick wall geometry). Although the data were based on laboratory result, the SVM model provides a helpful method in predicting cement-sheath-failure. This study provides a data based method to predict cement fatigue failure under cyclic changing pressure and temperature. The result will be instructive for the cement design and wellbore operation optimization.

A Deep Learning Model for Rate of Penetration Prediction and Drilling Performance Optimization Using Genetic Algorithm

Abstract: In this study, a deep learning model is proposed that can accurately predict the rate of penetration during geothermal or oil and gas well construction operations. Also, a genetic algorithm is applied and used together with the deep learning model to determine the optimum values for the drilling parameters: weight on bit (WOB) and drillstring rotation rate (RPM). It is vital to estimate the optimal set of values for drilling parameters to construct wellbores quickly and efficiently. Traditionally, drill-off tests are conducted by halting the normal drilling operation and manually changing the WOB and RPM values to search for the highest ROP output. This operation can be repetitive and can lead to an inaccurate estimation of parameters because only a few different parameters are tried. The proposed learning algorithm estimates the optimum WOB and RPM, based on the historical values and can keep learning as the drilling proceeds, which is essential for fully automated well construction. The proposed deep learning model is trained with actual drilling datasets that showed an accurate prediction of the rate of penetration and mechanical specific energy (MSE). This model is used together with the genetic algorithm and the optimum drilling parameters are determined that yield minimum MSE. The results showed a significant performance improvement compared to the historical values. The proposed model can be used as an advisory system to the driller or the output can be used within the control system to automate the drilling process. The proposed learning model showed the capability to further optimize the drilling-rate and mitigate any invisible lost time (ILT) and potential nonproductive time (NPT) by completing the well as soon as possible.

Application of Recurrent Neural Network Long Short-Term Memory Model on Early Kick Detection

Abstract: Long-short term memory [1] (LSTM) is an artificial Recurrent Neural Network (RNN) architecture capable of performing deep learning tasks. With the special feedback feature, the LSTM network is suitable for processing a sequence of data and making a sequence of predictions. It has been successfully applied to many disciplines such as speech recognition, language translation, time series forecasting, and anomaly detection. In this paper, the RNN-LSTM network is applied to real-time drilling data to study the complex dependencies between multiple drilling parameters and common kick indicators. A well-trained model will use the concept of the sliding window to continuously predict the unforeseen value of sensitive kick indicators. With proper analysis, the predicted result is helpful to detect kicks ahead of time. This paper also proposed a general workflow to easily visualize the prediction results. Compared with other time series prediction methods, the LSTM network has the advantages of more accurate multi-step prediction, more physical, and more flexible. The proposed LSTM network uses accelerated GPU computing, the fast computational speed makes both online and offline learning possible. It is concluded that this approach is capable of accurately predicting kick indicators under certain circumstances. It may provide innovative guidance for the application of the LSTM network in early kick detection and future study.