André Bulcão

Performance Evaluation of Optimized Implementations of Finite Difference Method for Wave Propagation Problems on GPU Architecture

The scattering of acoustic waves in non-homogeneous media has been of practical interest for the petroleum industry, mainly in the determination of new oil deposits. A family of computational models that represent this phenomenon is based on finite difference methods. The simulation of these phenomena demands a high computational cost. In this work we employ GPU for the development of solvers for a 2D wave propagation problem with finite difference methods. Although there are many related works that use the same implementation presented in this paper, we propose a detailed and novel performance and memory bottleneck analysis for this hardware architecture.

Target oriented illumination analysis using wave equation

The proposed methodology requires considerably less resources, in terms of processing time and disk space, than the traditional procedure using wave equation. The traditional procedure compares depth images obtained by a migration scheme, after modeling the complete dataset for a certain acquisition geometry. Both methodologies are more accurate than the results based on ray tracing modeling in complex areas (due to the high frequency approximations involved in this scheme).

An Approach to Optimise the Execution of RTM Algorithm in Multicore Machines

The new oil fields discovered in the Gulf of Mexico and in Brazils southeast coast are located in deep water, which imposes new challenges for sub salt seismic imaging. To produce sufficient accurate imaging of such fields, the compute intensive RTM method is the currently favoured approach, despite its high computational cost. This work evaluates the RTM code in multicore machines and proposes a new version that is more than 2.4 faster than the original. Besides, how the memory hierarchy is utilised has a crucial impact on the performance of the code. This article also presents an innovative approach to calculate efficient block values for a better memory utilisation on multicore archictetures.

Application of Blended Seismic Data in Full Waveform Inversion

Summary In seismic imaging the Full Waveform Inversion (FWI) has recently gained much importance, mainly due to advances in computational power and seismic acquisition improvements, especially in areas with high geological complexities. In fact, it can provide higher seismic resolution compared to pre-stack depth migration, which may contribute to better seismic interpretation. However, for complex geological models the success of seismic imaging depends on the design of the seismic experiment. In fact, besides the necessity of a wide spectrum of frequencies, it is required to properly register the scattering wavefields associated to the geological targets. Otherwise, the imaging and inversion schemes are not effective. Acquiring the scattering wavefields as a 3D field may be cost prohibitive using conventional surveys, where consecutive shot records don’t have time superposition. In this work, we abandon this condition, and extend the inversion scheme called Contrast Source Inversion (CSI) to handle data in which shots are superposed randomly (Blended data). On the propose method, the deblending phase (separation the interference between shots) is not necessary. Numerical results applied on Marmousi model show how efficient, accurate, and stable this method is when compared to results obtained with traditional acquisition approaches. Even in very low signal to noise ratio cases, except for small artifacts, it was possible to recover all features present in the original model.

A 3D Reverse Time Migration Scheme For Offshore Seismic Data

This work introduces a reverse time migration formulation specially designed for offshore data acquired with a single boat pulling one or two air-guns and some streamers of receivers. 3D depth migrations are performed using the concept of areal shot record, as introduced by Berkhout in 1992. In the proposed scheme, an areal shot record is computed through the sum of convolutions between the seismograms registered along a sail line and a synthesis operator, which is designed to form a specific wave front just above the target zone (controlled illumination). All extrapolations are performed solving the acoustic two-way wave equation without smooth the macro-velocity field. The method is illustrated with an application on the SEGEAGE Overthrust model. Except for some small artifacts, less than forty areal shot records were enough to image the main interfaces. The execution time was considerably less than that required to reserve time migrate all seismograms independently.

Qualitative and quantitative illumination analyses using wave equation

The proposed methodology is based on solutions for the Acoustic Wave Equation obtained by FDM (Finite Difference Method). The first step involves a wavefield extrapolation from the target up to the surface, applying the Reciprocity Principle like Alves et al. (2008). Thus a lower computational cost is obtained when compared to conventional methods, which make one extrapolation for the source and another extrapolation for the receiver.

Implementation Aspects of the 3D Wave Propagation in Semi-infinite Domains Using the Finite Difference Method on a GPU Based Cluster

The scattering of acoustic waves has been a matter of practical interest for the petroleum industry, mainly in the determination of new oil deposits. A family of computational models that represent this phenomenon is based on Finite Difference Methods (FDM). The simulation of these phenomena demands a high computational processing power and large amounts of memory. Furthermore, solving this problem in a high performance computing (HPC) environment requires the use of tools such as MPI (Message Passing Interface) and GPUs in order to soften the effort necessary on implementation. In this work a GPU based cluster environment is employed for the development of an efficient scalable solver for a 3D wave propagation problem using the FDM. The details related to the implementation of the FDM applied to wave propagation in GPUs are presented. A performance analysis for several simulations is also discussed. The solution discussed herein is suitable not only for a single GPU system, but for clusters of GPUs as well.

Numerical Experiments of Full Waveform Inversion on a Typical Pre-salt Model - Part 2 - Elastic Effects

We consider the elastic effects on an acoustic full waveform inversion generated by a synthetic pre-salt model of Santos basin, Brazil. We used P/S-velocity and density models to generate shot-gathers with acoustic, elastic (isotropic), constant and variable density modelling. The comparison of the amplitudes of the shot-gathers already provides a clue to the challenges acoustic FWI may face, as acoustic modelling is not able to provide the correct amplitude for the elastic data. This test is complemented with a constant-density acoustic FWI inversion, for all the different datasets, but with the same algorithm and parameterization. FWI with acoustic data was able to retrieve the correct information of high velocity sediments in the post/pre-salt environment and the stratification of the evaporitic sequence. On the other hand, the inversion of elastic data could not properly invert anything underneath the top of salt. The comparison of the inversion results indicates a direction in which algorithm and simplifications can be done to pursue an adequate property inversion.

Fault tolerance in an industrial seismic processing application for multicore clusters

Seismic processing applications are used to identify geological structures where reservoirs of oil and gas may be found. With oil companies seeking better precision over larger geographical regions, these applications require larger clusters to keep execution times reasonable. The combination of longer run times and clusters with greater numbers of components increases the probability of faults during the execution. To address this issue, this paper describes an application-level fault tolerance mechanism that considers node crashes and communication link failures. For this industrial application, experiments show that continued execution with the remaining resources is both feasible and efficient.

Contrast Source Inversion of Blended Data

Full waveform inversion has the potential of playing the central role in seismic imaging in the near future. In fact, its power of providing higher seismic resolution compared to pre-stack depth migration may contribute to seismic interpretation in areas which present high geological complexities. However, its success depends on the design of the seismic experiment. In fact, besides the necessity of a wide spectrum of frequencies, it is required to properly register the scattering wavefields associated to the geological targets. Otherwise, the inversion scheme is not effective. On the other hand, acquiring the scattering wavefields as a whole may be unfeasible using conventional surveys, where consecutive shot records don’t have time superposition. In this work, we abandon this condition, and extend the contrast source inversion method to handle data in which shots are superposed randomly. In our method, the deblending phase is not necessary. Applications on Marmousi model show how efficient, accurate, and stable this method is when compared to results obtained with traditional acquisition approaches. Even in very low signal to noise ratio cases, except for small artifacts, it was possible to recover all features present in the original model.