Comparison of permeability predictions on cemented sandstones with physics-based and machine learning approaches

Columbu, Stefano; Jensen, Jerry L.; Lake, Larry W.

doi:10.1016/j.jngse.2020.103244

Cited by 38 publications

(8 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Machine learning in general and deep learning in particular have seen a widespread utility from biomedical image segmentation [30] and manufacturing [31], to autonomous vehicles [32] and space [33] applications. A few recent studies have used machine learning as a regression technique in the form of neural networks trained over empirical and theoretical data [34], physics-based model data [35], and physics-plus-empirical data [36]. Time series forecasting is yet another area where machine learning has seen tremendous potential [37].…”

Section: Studies On Machine Learningmentioning

confidence: 99%

Lifetime Prediction Using a Tribology-Aware, Deep Learning-Based Digital Twin of Ball Bearing-Like Tribosystems in Oil and Gas

2021

View full text Add to dashboard Cite

The recent decline in crude oil prices due to global competition and COVID-19-related demand issues has highlighted the need for the efficient operation of an oil and gas plant. One such avenue is accurate predictions about the remaining useful life (RUL) of components used in oil and gas plants. A tribosystem is comprised of the surfaces in relative motion and the lubricant between them. Lubricant oils play a significant role in keeping any tribosystem such as bearings and gears working smoothly over the lifetime of the oil and gas plant. The lubricant oil needs replenishment from time to time to avoid component breakdown due to the increased presence of wear debris and friction between the sliding surfaces of bearings and gears. Traditionally, this oil change is carried out at pre-determined times. This paper explored the possibilities of employing machine learning to predict early failure behavior in sensor-instrumented tribosystems. Specifically, deep learning and tribological data obtained from sensors deployed on the components can provide more accurate predictions about the RUL of the tribosystem. This automated maintenance can improve the overall efficiency of the component. The present study aimed to develop a deep learning-based digital twin for accurately predicting the RUL of a tribosystem comprised of a ball bearing-like test apparatus, a four-ball tester, and lubricant oil. A commercial lubricant used in the offshore oil and gas components was tested for its extreme pressure performance, and its welding load was measured using a four-ball tester. Three accelerated deterioration tests was carried out on the four-ball tester at a load below the welding load. Based on the wear scar measurements obtained from the experimental tests, the RUL data were used to train a multivariate convolutional neural network (CNN). The training accuracy of the model was above 99%, and the testing accuracy was above 95%. This work involved the model-free learning prediction of the remaining useful lifetime of ball bearing-type contacts as a function of key sensor input data (i.e., load, friction, temperature). This model can be deployed for in-field tribological machine elements to trigger automated maintenance without explicitly measuring the wear phenomenon.

show abstract

Section: Studies On Machine Learningmentioning

confidence: 99%

Lifetime Prediction Using a Tribology-Aware, Deep Learning-Based Digital Twin of Ball Bearing-Like Tribosystems in Oil and Gas

2021

View full text Add to dashboard Cite

show abstract

“…Likewise, Al-Mudhafar employed a combination of machine learning and data analytics to enhance the geostatistical characterization of clastic reservoirs in the Luhais oil field. Male et al performed a comparative analysis between physics and machine-learning-based approaches for predicting permeability in cemented sandstones. Rafik and Kamel employed nonparametric regression with multivariate analysis and a neural network (NN) to predict permeability and porosity based on the well log data set.…”

Section: Introductionmentioning

confidence: 99%

Harnessing Advanced Machine-Learning Algorithms for Optimized Data Conditioning and Petrophysical Analysis of Heterogeneous, Thin Reservoirs

Manzoor

Ehsan

Hussain³

et al. 2023

Energy Fuels

View full text Add to dashboard Cite

Petrophysical analysis is an industry-standard practice for reservoir evaluation as it provides critical inputs for characterizing subsurface formations and estimating resource potential. Khadro/Ranikot Formation sands are proliferous producers in the Central Indus Basin, Pakistan. The demarcate potential in intercalated sand shale layers that are thin and heterogeneous makes it a challenging reservoir. Conventional petrophysical interpretation is laborious and does not produce upto-mark results due to reservoir complexity, data limitations, and associated uncertainties. Hence, an emerging and delicate machinelearning (ML) approach has been comprehensively applied to analyze the potential and robustly interpret well log data while addressing the associated challenges. This case study entails a thorough evaluation of well log quality, assessing several algorithms such as least-squares support vector machines (one-class SVM), Random Forest Regressor (RFR), Extra Tree Regressor (ETR), Gradient Boosting Regressor (GBR), Decision Tree Classifier (DTC), etc. to compare their efficacy and reliability. One-class SVM helps to reduce outliers with great certainty, while the missing logs sonic (DT) and density (RHOB) are precisely predicted via GBR and ETR with 0.66 and 0.88 R 2 , respectively. Hence, providing reliable and optimized quality logs suitable for ML-based petrophysics. ML worked on these augmented logs by dividing the data into 60% training and 40% testing. The ETR outperformed the rest of the models with a correlation of 0.99 and 0.91 among conventional and ML results. Likewise, RFR performed exceptionally well for water saturation modeling, expressing the highest 0.93 correlation. Finally, DTC modeled reservoir facies with the best 91% accuracy and 0.935 F1 measures at the blind well. Excellent calibration of >85% is met with the estimates obtained by the predictive model compared to conventional methods. This comprehensive approach offers cost-effective and robust workarounds for modern formation evaluation with minimal uncertainty and resource-efficient multiwell interpretation within complex reservoirs and sets the stage for further research in the machine-learning ecosystem.

show abstract

“…In general, the main physics-based models used for predicting the permeability include index model, Kozeny-Carman model, Timur model, and Herron model (Male et al 2020;Huang et al 2021). In these models, the permeability is primarily seen as a function of effective porosity, with the function developed using the results of experimental works on a set of cores.…”

Section: Introductionmentioning

confidence: 99%

Prediction of permeability of highly heterogeneous hydrocarbon reservoir from conventional petrophysical logs using optimized data-driven algorithms

Sheykhinasab

Mohseni

Bahari³

et al. 2022

J Petrol Explor Prod Technol

View full text Add to dashboard Cite

Permeability is an important parameter in the petrophysical study of a reservoir and serves as a key tool in the development of an oilfield. This is while its prediction, especially in carbonate reservoirs with their relatively lower levels of permeability compared to sandstone reservoirs, is a complicated task as it has larger contributions from heterogeneously distributed vugs and fractures. In this respect, the present research uses the data from two wells (well A for modeling and well B for assessing the generalizability of the developed models) drilled into a carbonate reservoir to estimate the permeability using composite formulations based on least square support vector machine (LSSVM) and multilayer extreme learning machine (MELM) coupled with the so-called cuckoo optimization algorithm (COA), particle swarm optimization (PSO), and genetic algorithm (GA). We further used simple forms of convolutional neural network (CNN) and LSSVM for the sake of comparison. To this end, firstly, the Tukey method was applied to identify and remove the outliers from modeling data. In the next step, the second version of the nondominated sorting genetic algorithm (NSGA-II) was applied to the training data (70% of the entire dataset, selected randomly) to select an optimal group of features that most affect the permeability. The results indicated that although including more input parameters in the modeling added to the resultant coefficient of determination (R2) while reducing the error successively, yet the slope of the latter reduction got much slow as the number of input parameters exceeded 4. In this respect, petrophysical logs of P-wave travel time, bulk density, neutron porosity, and formation resistivity were identified as the most effective parameters for estimating the permeability. Evaluation of the results of permeability modeling based on root-mean-square error (RMSE) and R2 shed light on the MELM-COA as the best-performing model in the training and testing stages, as indicated by (RMSE = 0.5600 mD, R2 = 0.9931) and (RMSE = 0.6019 mD, R2 = 0.9919), respectively. The generalizability assessment conducted on the prediction of permeability in well B confirmed the MELM-COA can provide reliable permeability predictions by achieving an RMSE of 0.9219 mD. Consequently, the mentioned methodology is strongly recommended for predicting the permeability with high accuracy in similar depth intervals at other wells in the same field should the required dataset be available.

show abstract

Comparison of permeability predictions on cemented sandstones with physics-based and machine learning approaches

Cited by 38 publications

References 23 publications

Lifetime Prediction Using a Tribology-Aware, Deep Learning-Based Digital Twin of Ball Bearing-Like Tribosystems in Oil and Gas

Lifetime Prediction Using a Tribology-Aware, Deep Learning-Based Digital Twin of Ball Bearing-Like Tribosystems in Oil and Gas

Harnessing Advanced Machine-Learning Algorithms for Optimized Data Conditioning and Petrophysical Analysis of Heterogeneous, Thin Reservoirs

Prediction of permeability of highly heterogeneous hydrocarbon reservoir from conventional petrophysical logs using optimized data-driven algorithms

Contact Info

Product

Resources

About