FAULT DELINEATION ON SESMIC POST-STACK 3D DATA USING ARTIFICIAL INTELLIGENCE

• Main Author: Febriarta, Bondan
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/76038
• Institution: Institut Teknologi Bandung

The study of fault delineation in 3D post-stack seismic data with the implementation of the convolutional neural network algorithm will be carried out using the U-Net 3D architecture. The purpose of this study is to build a CNN architecture and compare the results of fault requests with attribute variants. The method used in developing this fault program delineation is CNN with the Pytorch core library. The dataset was prepared by utilizing 220 pairs of synthetic data consisting of 200 pairs of train/test data and 20 pairs of validation data. The results obtained in the training process show that the loss function curve has converged, which is around below 0.0154 for train and around 0.0308 for test, where this convergence indicates the success of the training process. Estimates of the fault delineation estimation using the CNN model show good values based on the performance metrics used, namely precision, recall, and f-1 scores using validation data, which are around 0.7, 0.8, and 0.9, respectively. Qualitatively or visually, the estimation of fault delineation results on synthetic validation data using the CNN model outperforms the attribute variance, where the estimation of the fault attribute variance produced is not continuous and there are still many points that miss the estimate compared to the CNN model. For a while the delineation estimates often outperform the visual attribute variations. Likewise when the CNN model that has gone through the training process is applied to some field data. The results achieved when seen visually are good, where the areas where there are faults can be seen clearly, but in areas where the faults coincide with each other, the resulting delineation is not accurate.