Penggunaan Deep Learning Untuk Estimasi Life Time Trafo Pada Gardu Induk PT. Petrokimia Gresik

Julianto, Wahid (2026) Penggunaan Deep Learning Untuk Estimasi Life Time Trafo Pada Gardu Induk PT. Petrokimia Gresik. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022241111-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Keandalan transformator daya merupakan pilar fundamental dalam menjaga stabilitas sistem tenaga listrik dan menunjang kontinuitas proses produksi industri skala masif di PT Petrokimia Gresik. Secara konvensional, pemantauan kondisi isolasi trafo dievaluasi menggunakan Health Index (HI). Namun, metode evaluasi statis ini rentan menghasilkan diagnosis yang bias (false-positive) apabila terdapat parameter uji diagnostik lapangan yang tidak lengkap, seperti absennya hasil uji kelarutan senyawa furan atau pengujian properti dielektrik minyak. Untuk mengatasi batasan operasional tersebut, penelitian ini mengusulkan kerangka kerja pemodelan Deep Learning berbasis arsitektur Long Short-Term Memory (LSTM) yang dirancang agar adaptif terhadap variasi ketersediaan data tanpa mengorbankan integritas evaluasi aset. Memanfaatkan dataset historis deret waktu (time-series) dari 27 unit transformator daya, model LSTM dengan konfigurasi hidden layer ganda (64 dan 32 neuron) dilatih secara sekuensial untuk menangkap memori jangka panjang dari pola degradasi aset. Hasil pengujian empiris membuktikan bahwa arsitektur jaringan ini memiliki kapabilitas tinggi untuk mengekstrapolasi Remaining Useful Life (RUL) dan memproyeksikan lintasan penurunan kesehatan trafo hingga 24 bulan ke depan. Model prediktif ini tervalidasi dengan tingkat galat yang sangat presisi, mencatatkan Mean Absolute Error (MAE) sebesar 3,69% dan Root Mean Squared Error (RMSE) sebesar 5,20%. Kerangka kerja ini difungsikan sebagai Early Warning System (EWS) yang mumpuni, memungkinkan departemen Maintenance untuk bertransisi dari Time-Based Maintenance menjadi Condition-Based Maintenance (perawatan prediktif berbasis kondisi), guna memitigasi kegagalan katastrofik secara proaktif.
==================================================================================================================================
The operational reliability of power transformers is a fundamental pillar in maintaining electrical power system stability and ensuring the continuity of large-scale industrial production at PT Petrokimia Gresik. Conventionally, transformer insulation condition monitoring is evaluated using the Health Index (HI). However, this static evaluation method is highly susceptible to diagnostic bias (false-positives) when field diagnostic parameters are incomplete, such as due to the absence of furan analysis or insulating oil dielectric property tests. To overcome these operational constraints, this study proposes a deep learning framework based on the Long Short-Term Memory (LSTM) architecture, designed to be adaptive to variations in data availability without compromising the integrity of asset evaluation. Leveraging a historical time-series dataset from 27 power transformer units, an LSTM model with a dual-hidden-layer configuration (64 and 32 neurons) was sequentially trained to capture the long-term dependencies of asset degradation patterns. Empirical results demonstrate that this network architecture possesses a high capability to extrapolate the Remaining Useful Life (RUL) and project transformer health degradation trajectories up to 24 months into the future. The predictive model is validated with exceptional precision, achieving a Mean Absolute Error (MAE) of 3.69% and a Root Mean Squared Error (RMSE) of 5.20%. This framework functions as a robust Early Warning System (EWS), enabling the Maintenance department to transition from rigid Time-Based Maintenance to predictive Condition-Based Maintenance (CBM), thereby proactively mitigating catastrophic failures.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Transformator Daya, Long Short-Term Memory, Health Index, Remaining Useful Life, Condition-Based Maintenance. Power Transformer, Long Short-Term Memory, Health Index, Remaining Useful Life, Condition-Based Maintenance.
Subjects:	Q Science Q Science > Q Science (General) > Q325 GMDH algorithms. Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Wahid Julianto
Date Deposited:	31 Jul 2026 07:07
Last Modified:	31 Jul 2026 07:07
URI:	http://repository.its.ac.id/id/eprint/140854

Actions (login required)

View Item