The pristine filter papers were crosslinked with polyvinyl alcohol by tolylene diisocyanate, leading to a superoleophobic surface which enables excellent oil–water separation efficiency.
Automating geobodies using insufficient labeled training data as input for structural prediction may result in missing important features and a possibility of overfitting, leading to low accuracy. We adopt a deep learning (DL) predictive modeling scheme to alleviate detection of channelized features based on classified seismic attributes (X) and different ground truth scenarios (y), to imitate actual human interpreters’ tasks. In this approach, diverse augmentation method was applied to increase the accuracy of the model after we were satisfied with the refined annotated ground truth dataset. We evaluated the effect of dropout as a training regularizer and facies’ spatial representation towards optimized prediction results, apart from conventional hyperparameter tuning. From our findings, increasing batch size helps speedup training speed and improve performance stability. Finally, we demonstrate that the designed Convolutional Neural Network (CNN) is capable of learning channelized variation from complex deepwater settings in a fluvial-dominated depositional environment while producing outstanding mean Intersection of Union (IoU) (95%) despite utilizing 6.4% from the overall dataset and avoiding overfitting possibilities.
The prediction of subsurface properties such as velocity, density, porosity, and water saturation has been the main focus of petroleum geosciences. Advanced methods such as Full Waveform Inversion (FWI), Joint Migration Inversion (JMI) and ML-Rock Physics are able to produce better predictions than their predecessors, but they still require tedious manual interpretation that is prone to human error. The research on these methods remains open as they suffer from technical limitations. As computing resources are becoming cheaper, the use of a single deep-generative adversarial network is feasible in predicting all these properties in a completely data-driven manner. In our proposed method of multiscale pix2pix applied to SEG SEAM salt data, we have managed to map from one input, which is seismic post-stack data, to several outputs of reservoir and elastic properties such as porosity, velocity, and density by using only one trained model and without having to manually interpret or pre-process the input data. With 90% accuracy of the results in the synthetic data testing, the method is worthy of being explored by the petroleum geoscience fraternity.
Objectives/Scope The growth of seismic attributes technology in accelerating the advancement of exploration and development started in late 1960s, with identification of bright spots through seismic reflection traits. In between the captured timeline, pattern recognition or neural network-based analysis on seismic attributes since 1990s gradually bring significant improvement in defining structural or depositional environment. Dunham (2020) addressed insufficient labelled geophysical data for supervised classification challenges lead to the possibility of overtraining for small-scale labelled datasets and unstable predictions for unlabeled datasets. Automated 3D interpretation using CNN produced less convincing accuracy on dataset with larger incised channel (Gao, 2021). We explore an approach to improve channelized features prediction accuracy through Deep Learning (DL) algorithms by refining the quality of the input dataset. Methods, Procedures, Process We developed supervised learning method using pixel-based segmentation and transformed matrix data into structural interpretation by constraining multi seismic attributes with ground truth dataset. It is divided into two (2) steps: 1a) Exploratory Data Analysis (EDA) where sixteen (16) seismic attributes representing categories defined by Roden (2015) were extracted and normalized for feature's correlation; 1b) We choose fluvial dominated field dataset located in Malay Basin and use Python-based Labelling Tool (PyLT) to annotate 64 time slices penetrating I-27 to I-35 Lower Reservoir of Mid Miocene. Data cube were 2 ms-sampled with size of 1024 × 256 (inlines, crosslines) annotated in z-direction. Associated fluvial facies elements were interpreted into five (5) classes; 2) Application of DL in data augmentation and conducted hyperparameter testing to find the network configuration for optimal performance. Results, Observations, Conclusions A range of learning rates have been tested, from 1e-4 to 5e-6 at increments of 5e-6. We have settled on a learning rate of 5e-5, which provided the fastest training time without adversely affecting the training process by introducing unwanted instabilities. Small weight decay multiplier of 1e-3 is used to introduce weight regularization to mitigate overfitting. We tested various batch sizes from 8 to 128, in increments of powers of two. From our observation, increasing batch size led to increases in training speed and improved training stability. We have applied various range of drop out to prevent overfitting by removing unnecessary features during training stage and stabilized predictions while being the regularizer. It was observed dropout slows down the learning rate but yielded performance improvements on test dataset when increased from 0 to 0.3. Figure 1 (a) portrays two (2) inline sections representing spectral decomposition volumes with Log Pattern Analysis from nearby key wells where we can observe: Figure 1 Integrated methodology applied in this study consists of four (4) main steps: 1. Exploratory Data Analysis (EDA) for input data; 2. Define the ground truth based on two (2) scenarios; 3. Training the dataset and monitor the model's performance; 4. Geobodies extraction to be integrated with other relevant dataset. Figure 1 (b – input column), seismic attributes example (inset shown the Sweetness attributes) as part of the input data. We have chosen two (2) different time slices penetrating the studied channelized reservoir interval to identify different fluvial structures. Figure 1b (ground truth column) depicted dataset built using labelling tool. (Figure 1b – Prediction's column) shows It was observed that neural network is capable to capture distributary channels missed by attributes in binary classifications and Mud-Filled Channel (MFC) still can be detected through multi-facies classification even though facing small training dataset in Class III, as observed in deeper slice. Novel/Additive Information We have adopted DL method to alleviate detection of fluvial facies despite dealing with limited dataset. Diverse augmentation method was applied to increase the accuracy of model after we have satisfied with the refined annotated ground truth dataset. Random flipping and rotations from 0° to 45° were applied to all tiles and corresponding labels to prevent possibilities of data overfitting and helps to generalize potential aspects of features that DL algorithms have not seen before.
A couple SPH-FEA modelling technique is investigated to determine the suitability of the mentioned technique to model offshore cutting operation of Abrasive Water-jet (AWJ) cutting. For model validation, results are compared to analysis available in literature on abrasive water-jet machining. An AWJ cutting a ductile metal sample with certain thickness and property is modeled and simulated to investigate the impact response and parameters that influence, including impact speeds. To overcome the difficulties of fluid—solid interaction and extra-large deformation problem using finite element method (FEM), the SPH-coupled FEA modeling for abrasive waterjet cutting simulation is presented, in which the abrasive waterjet is modeled by SPH particles and the target material is modeled by Finite Element. Validation is achieved by comparing the model against numerical result available in literature. The coupled SPH-FEA model is compared against previously published results and validated. Then, studies is conducted to determine the erosion rate effect due to change in abrasive types and different flow rates. The depth of penetrations and erosion rate is extracted for analysis. It can be seen that the generated SPH-FEA method is able to simulate the condition of AWJ cutting of offshore steel structure material up to 250 MPa. The fluid-structure interaction in AWJ cutting where the erosion-cutting of steel target material is highly non-linear was modelled by full coupled SPH/FEA. This research demonstrates that the approach can be extended to full-scale AWJ-steel structure cutting simulation through appropriate management of SPH/FEA resources. With a validated model, researchers will be able to manipulate the variables in AWJ to study the efficacy of cutting optimization in offshore decommissioning operations. The novelty in this paper is that the coupled SPH-FEA technique to model Abrasive Water Jetting is used to model cutting of offshore structures. This validated coupled SPH-FEA modelling technique will enable engineers to design new cutting tools that can improve efficiency and efficacy of next generation offshore cutting tools.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
customersupport@researchsolutions.com
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.