Autopicking Fault Menggunakan Fault Semi-Supervised Learning Untuk Identifikasi Patahan Di Jawa Timur

Ilham, Syamsul Arifin (2026) Autopicking Fault Menggunakan Fault Semi-Supervised Learning Untuk Identifikasi Patahan Di Jawa Timur. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5017221029-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Interpretasi patahan pada data seismik refleksi merupakan tahapan penting dalam memahami struktur bawah permukaan, terutama pada kegiatan eksplorasi hidrokarbon dan studi tektonik regional. Namun, interpretasi manual memiliki keterbatasan berupa subjektivitas, membutuhkan waktu yang relatif lama, serta sulit diterapkan secara konsisten pada data berskala besar. Penelitian ini bertujuan untuk mengevaluasi kemampuan metode Fault Semi-Supervised Learning (FaultSSL) dalam melakukan autopicking fault pada data seismik refleksi 2D di Jawa Timur serta menganalisis pengaruh penggunaan atribut seismik terhadap hasil deteksi patahan. Metode FaultSSL dibangun menggunakan framework Mean Teacher yang terdiri atas student network dan teacher network dengan backbone High Resolution Network (HRNet). Proses pelatihan memanfaatkan mekanisme Exponential Moving Average (EMA), PaNning Consistency (PNC), dan PaTching Consistency (PTC), sedangkan fungsi loss yang digunakan adalah Mask Dice Loss (MD Loss) dan Cosine Loss. Setelah pelatihan selesai, model digunakan untuk menghasilkan fault probability map yang kemudian dikonversi menjadi kandidat patahan melalui proses thresholding. Selain menggunakan data seismik asli, penelitian ini juga menguji pengaruh sembilan atribut seismik, yaitu Coherence, Variance, Similarity, Chaos, Dip Deviation, Instantaneous Phase, Energy Gradient, Curvature, dan Ant Tracking terhadap kualitas hasil autopicking. Hasil penelitian menunjukkan bahwa FaultSSL mampu mendeteksi sebagian besar patahan pada data seismik dengan kualitas yang baik, ditunjukkan oleh dominasi prediksi true positive meskipun masih ditemukan beberapa false positive terutama pada area batas window. Pada data dengan kualitas lebih rendah, model tetap menghasilkan interpretasi yang relatif konsisten dengan kecenderungan false negative yang sesuai dengan karakteristik data. Analisis atribut seismik menunjukkan bahwa setiap atribut memberikan respons yang berbeda terhadap proses deteksi patahan. Kelompok atribut Coherence, Similarity, Chaos, dan Ant Tracking cenderung mengikuti pola horizon sehingga berpotensi menyebabkan under-picking, sedangkan Dip Deviation, Curvature, dan Energy Gradient lebih sensitif terhadap perubahan geometri reflektor sehingga mampu menonjolkan pola sesar dengan lebih baik walaupun menghasilkan kecenderungan over-picking. Berdasarkan hasil penelitian dapat disimpulkan bahwa FaultSSL memiliki kemampuan yang baik dalam mendeteksi patahan pada data seismik refleksi 2D dan berpotensi digunakan sebagai alat bantu interpretasi struktur geologi secara otomatis, sementara atribut Dip Deviation, Curvature, dan Energy Gradient merupakan atribut yang paling efektif dalam mendukung proses autopicking fault pada penelitian ini.
================================================================================================================================
Fault interpretation on seismic reflection data is an essential stage in understanding subsurface structures, particularly in hydrocarbon exploration and regional tectonic studies. However, manual interpretation is often time-consuming, subjective, and difficult to apply consistently to large datasets. This study aims to evaluate the capability of the Fault Semi-Supervised Learning (FaultSSL) method for automatic fault picking on 2D seismic reflection data from East Java and to investigate the influence of seismic attributes on fault detection results. FaultSSL is built upon a Mean Teacher framework consisting of a student network and a teacher network with a High-Resolution Network (HRNet) backbone. The training process employs Exponential Moving Average (EMA), PaNning Consistency (PNC), and PaTching Consistency (PTC), while Mask Dice Loss (MD Loss) and Cosine Loss are used as the objective functions. After training, the model generates a fault probability map, which is subsequently converted into fault candidates through a thresholding process. In addition to the original seismic data, nine seismic attributes were evaluated, namely Coherence, Variance, Similarity, Chaos, Dip Deviation, Instantaneous Phase, Energy Gradient, Curvature, and Ant Tracking, to assess their impact on automatic fault detection performance. The results demonstrate that FaultSSL successfully identifies most faults in high-quality seismic data, as indicated by the predominance of true-positive predictions, although several false positives are still observed, particularly near prediction window boundaries. For lower-quality seismic data, the model maintains relatively consistent interpretations, with false negatives mainly occurring in areas characterized by poor reflector continuity and low signal quality. The attribute analysis reveals that each seismic attribute responds differently to fault detection. Coherence, Similarity, Chaos, and Ant Tracking tend to follow seismic horizons, leading to under-picking in certain fault zones, whereas Dip Deviation, Curvature, and Energy Gradient are more sensitive to reflector geometry changes and are therefore more effective in highlighting fault patterns, although they also increase the tendency for over-picking and false-positive detections. Based on these findings, FaultSSL demonstrates strong potential as an automated structural interpretation tool for 2D seismic reflection data, while Dip Deviation, Curvature, and Energy Gradient are identified as the most effective attributes for supporting automatic fault picking in this study.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Autopicking Fault, FaultSSL, Mean Teacher, Atribut Seismik
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Civil, Environmental, and Geo Engineering > Geophysics Engineering > 33201-(S1) Undergraduate Theses
Depositing User:	Syamsul Arifin Ilham
Date Deposited:	27 Jul 2026 00:45
Last Modified:	27 Jul 2026 00:45
URI:	http://repository.its.ac.id/id/eprint/137207

Actions (login required)

View Item