Machine learning methods applied to drilling rate of penetration prediction and optimization - A review

Drilling rate of penetration prediction and optimization using response surface methodology and bat algorithm

Development of New Permeability Formulation From Well Log Data Using Artificial Intelligence Approaches

The rheological properties of the drilling fluid play a key role in controlling the drilling operation. Knowledge of drilling fluid rheological properties is very crucial for drilling hydraulic calculations required for hole cleaning optimization. Measuring the rheological properties during drilling sometimes is a time-consuming process. Wrong estimation of these properties may lead to many problems, such as pipe sticking, loss of circulation, and/or well control issues. The aforementioned problems increase the non-productive time and the overall cost of the drilling operations. In this paper, the frequent drilling fluid measurements (mud density, Marsh funnel viscosity (MFV), and solid percent) are used to estimate the rheological properties of bentonite spud mud. Artificial neural network (ANN) technique was combined with the self-adaptive differential evolution algorithm (SaDe) to develop an optimum ANN model for each rheological property using 1029 data points. The SaDe helped to optimize the best combination of parameters for the ANN models. For the first time, based on the developed ANN models, empirical equations are extracted for each rheological parameter. The ANN models predicted the rheological properties from the mud density, MFV, and solid percent with high accuracy (average absolute percentage error (AAPE) less than 5% and correlation coefficient higher than 95%). The developed apparent viscosity model was compared with the available models in the literature using the unseen dataset. The SaDe-ANN model outperformed the other models which overestimated the apparent viscosity of the spud drilling fluid. The developed models will help drilling engineers to predict the rheological properties every 15–20 min. This will help to optimize hole cleaning and avoid pipe sticking and loss of circulation where bentonite spud mud is used. No additional equipment or special software is required for applying the new method.

Real Time Determination of Rheological Properties of Spud Drilling Fluids Using a Hybrid Artificial Intelligence Technique

Predictive models have been widely used in different engineering fields, as well as in petroleum engineering. Due to the development of high-performance computer systems, the accuracy and complexity of predictive models have been increased significantly. One of the common methods for prediction is artificial neural network (ANN). ANN models in combination with optimization algorithms provide a powerful and fast tool for the prediction and optimization of processes which take a large amount of time if they are simulated using common simulation technics. In the present paper, to predict penetration rate during drilling process, several ANN models were developed based on the data obtained from drilling of a gas well located in south of Iran. Regarding the R2 and RMSE values of the developed models, the best model was selected for prediction of penetration rate. In the next step, artificial bee colony algorithm was used for optimization of the parameters which are effective on rate of penetration (ROP). Results showed that the model is accurate enough for being used in the prediction and optimization of ROP in drilling operations.

A new methodology for optimization and prediction of rate of penetration during drilling operations

Real-time drilling optimization improves drilling performance by providing early warnings in operation Mud hydraulics is a key aspect of drilling that can be optimized by access to real-time data. Different from the investigated references, reliable prediction of pump pressure provides an early warning of circulation problems, washout, lost circulation, underground blowout, and kicks. This will help the driller to make necessary corrections to mitigate potential problems. In this study, an artificial neural network (ANN) model to predict hydraulics was implemented through the fitting tool of matlab. Following the determination of the optimum model, the sensitivity analysis of input parameters on the created model was investigated by using forward regression method. Next, the remaining data from the selected well samples was applied for simulation to verify the quality of the developed model. The novelty is this paper is validation of computer models with actual field data collected from an operator in LA. The simulation result was promising as compared with collected field data. This model can accurately predict pump pressure versus depth in analogous formations. The result of this work shows the potential of the approach developed in this work based on NN models for predicting real-time drilling hydraulics.

Application of Real-Time Field Data to Optimize Drilling Hydraulics Using Neural Network Approach

Drilling Hydraulics Optimization Using Neural Networks


 As advanced drilling technology gains more popularity in the development of unconventional reservoirs and deepwater fields, the demand for an improved drilling hydraulics modeling escalates today. The situation is also true for foam applications that have long been used as a useful means of downhole cleaning and underbalanced drilling methods.
 This study investigates how a new foam model, recently developed by Wang et al. (2017), can be applied to a wide range of hole cleaning and drilling scenarios. The model combines rheological properties of wet foams and dry foams by using 9 model parameters (3 to define the range of gas and liquid flowrates and corresponding frictional pressure losses of interest, 4 to fit the power-law rheology of wet and dry foams, and last 2 to capture the sensitivity of foam rheology to gas and liquid rates). These scenarios consider foam circulation into 10,000 ft long wells at different inclination angles with a long vertical, inclined, or horizontal trajectory. The results from this new method are compared with two existing foam modeling techniques, so-called Chen et al.'s model (based on the correlations for wet foams only) and Edrisi and Kam's model (based on wet- and dry-foam rheological properties with 5 model parameters). The results show that, with or without formation fluid influx, the new foam model demonstrates the robustness of the new modeling technique in all scenarios capturing foam flow characteristics better, whenever the situation forms stable fine-textured foams or unstable coarse-textured foams.

Modeling of Foam-Assisted Wellbore Cleanup and Drilling Processes with Both Dry- and Wet-Foam Rheological Properties


 Although leak incidents continue, a pipeline remains the most reliable mode of transportation within the oil and gas industry. It becomes even more important today because the projection for new pipelines is expected to increase by 1 billion BOE through 2035. In addition, increasing number and length of subsea tiebacks face new challenges in term of data acquisition, monitoring, analysis, and remedial actions. Passive leak-detection methods commonly used in the industry have been successful with some limitations in that they often cannot detect small leaks and seeps. In addition to a thorough review of related topics, this study investigates how to create a framework for a smart pigging technique for pipeline leak detection, as an active leak detection method.
 Numerical modeling of smart pigging for leak detection requires two crucial components: detailed mathematical descriptions for fluid-solid and solid-solid interactions around pig, and network modeling for the calculation of pressure and rate along the pipeline using iterative algorithms. The first step of this study is to build a numerical model that shows the motion of a pig along the pipeline with no leak, i.e., at a given injection rate, a pig first accelerates until it reaches its terminal velocity, beyond which the pig moves at a constant velocity. The second step is to construct a network model that consists of two pipeline segments (one upstream and the other downstream of leak location) through which the pig travels and at the junction of which fluid leak occurs. By putting these multiple mechanisms together and using resulting pressure signatures, this study presents a new method to predict the location and size of a leak present in pipeline.

Yanfang Wang

Papers

Application of Real-Time Field Data to Optimize Drilling Hydraulics Using Neural Network Approach

Drilling Hydraulics Optimization Using Neural Networks

Modeling of Foam-Assisted Wellbore Cleanup and Drilling Processes with Both Dry- and Wet-Foam Rheological Properties

Modeling of Smart Pigging for Pipeline Leak Detection: From Mathematical Formulation to Large-Scale Application

Evaluation of Software-based Early Leak Warning System in Gulf-of-Mexico Subsea Flowlines