Application of Neural Networks for Predictive Control in Drilling Dynamics

Abstract

Real-time monitoring of BHA and drill bit dynamic behavior is a critical factor in improving drilling efficiency. It allows the driller to avoid detrimental drillstring vibrations and maintain optimum drilling conditions through periodic adjustments to various surface control parameters (such as hook load, RPM, flow rate and mud properties). However, selection of the correct control parameters is not a trivial task. A few iterations in parameter modification may be required before the desired effect is achieved and, even then, the result may not be optimal. For this reason, the development of efficient methods to predict the dynamic behavior of the BHA, and methods to select the appropriate control parameters, is important for drilling optimization. The approach presented in this paper uses the power of Neural Networks (NN) to model the dynamic behavior of the non-linear, multi- input/output drilling system. Such a model, along with an optimizing controller, provides the driller with a quantified recommendation on the appropriate corrective action(s) required to bring the system to an optimal drilling condition. Development of the NN model used drilling dynamics data from a field test. This field test involved various drilling scenarios in different lithologic units. The training and fine-tuning of the basic model utilized both surface and downhole dynamics data recorded in real-time while drilling. Measurement of the dynamic state of the BHA was achieved using data from downhole vibration sensors. This information, which represents the effects of modifying the surface control parameters, was recorded in the memory of the downhole tool. Representative portions of this test data set, along with the corresponding set of input-output control parameters, were used in developing and training the model. Test results are promising: There is good agreement between the dynamic behavior of the BHA predicted by the NN model and the actual measured BHA response. In addition, the test established criteria for selecting the most important input-output parameters and for selecting representative data sets for building and training the model. This analysis has demonstrated a promising approach to simulation and prediction of the dynamic behavior of the complex multi-parameter drilling system. This method could become a powerful alternative to traditional analytic or direct numerical modeling and its utilization could be extended beyond drilling dynamics to the field of drilling control and optimization.

Artificial neural network models to predict lost circulation in natural and induced fractures

“…This is done to ensure the model's robustness at generalizing to new data. Training data are used in training, verification data are Arehart [18] Drill bit diagnosis Used ANNs to find the state of wear of drill bit during drilling Dashevskiy et al [19] Real-time drilling dynamic Modeling the dynamic behavior of drilling system Bilgesu et al [20] Drill bit selection Used ANNs to select the "best" bit based on some inputs Ozbayoglu et al [21] Bed height for horizontal wells Used ANNs to predict bed heights in horizontal or highly inclined wellbores Vassallo et al [22] Bit bounce detection Used ANNs to detect bit bounce that can be used as a proactive approach to prevent bit whirl and stick-slip Fruhwirth et al [23],Wang and Salehi [24] Drilling hydraulics optimization and prediction Used ANNs to optimize and predict drilling hydraulics with a practical example Moran et al [25],Al-AbdulJabbar et al [26] Rate of penetration (ROP) prediction Used ANNs to predict ROP so that the drill time can be estimated better Gidh et al [27] Bit wear prediction Used ANNs to predict/ manage bit wear to improve ROP Lind & Kabirova [28] Drilling troubles prediction Used ANNs to forecast problems during the drilling process Okpo et al [29] Wellbore instability Wellbore stability prediction Ahmadi et al [30] Prediction of mud weight at wellbore conditions…”

Risk Factors Evaluation for Monitoring of Well Drilling

“…Another application of the neural network was found by Dashevskiy and others [43]. This work allowed to simulate a nonlinear drilling system with the minimal error share by monitoring its dynamic behavior.…”

A Game-Theoretic Approach to Improve Energy-Related Data

“…Drilling is one of these energy-related operations that produce a massive amount of data. As a result, multiple drilling-related AI models were developed such as stuck pipe detection, BHA monitoring, tool wear and flank wear analysis, , rate of penetration (ROP) prediction, formation top prediction, and much more. But due to the operations’ tough environment, a substantial amount of these data sets was not accurate. , As a result, developed AI models could not be fully upscaled the real-time operations; the bottleneck was the data quality.…”