Model Prediksi Fault And Lifetime Condition Pada Trafo Daya Di Industri Kelistrikan

Fauzi, M Annas Albab (2024) Model Prediksi Fault And Lifetime Condition Pada Trafo Daya Di Industri Kelistrikan. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6010221015-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (4MB) | Request a copy

Abstract

Manajemen aset yang efektif untuk trafo daya memerlukan data rinci dan analisis komprehensif untuk memperkirakan kondisi dan sisa masa pakainya. Penelitian ini bertujuan mengembangkan model prediksi fault dan lifetime menggunakan data Analisis Gas Terlarut (DGA). Empat metode yang digunakan mencakup Metode Rasio Doernenburg (DRM), Metode Rasio Roger (RRM), Metode Rasio IEC (IRM), Metode Segitiga Duval (DTM), dan penilaian derajat polimerisasi isolasi kertas dengan nilai furan.
Hasil penelitian menunjukkan algoritma Random Forest memiliki kinerja terbaik dalam memprediksi kondisi trafo pada model prediksi fault dan lifetime. Korelasi pada model lifetime antara CO2 dan furan mencapai 0,32, total combustible 0,19, CO 0,15, CO2/CO 0,12, dan asetilena 0,12. Akurasi model prediksi fault adalah DRM (47%), RRM (78%), IRM (89%), DTM (96%), dan DTA (98%). Model lifetime berbasis Random Forest dengan hyperparameter menghasilkan MAE sebesar 38,41 (ppb) dan MAPE sebesar 5,35%.
Pengujian data menggunakan lima metode (DRM, RRM, IRM, DTM, dan DTA) mencapai akurasi 98,2%, mengungguli pendekatan empat metode (92%). Distribusi kondisi transformator menunjukkan kondisi khusus (42%), Tingkat 1: Kritis (11%), Tingkat 2: Risiko Tinggi (31%), Tingkat 3: Risiko Sedang (15%), Tingkat 4: Risiko Rendah (2%), dan Tingkat 5: Risiko Minimal (0%). Trafo dengan ID 49, 50, 47, 34, 33, dan 32 memerlukan tindakan segera.
Studi ini menegaskan manfaat signifikan dari model prediksi berbasis machine learning dalam manajemen transformator daya. Pengembangan lanjutan melibatkan penambahan data historis, integrasi model dengan sistem pemantauan digital, pelatihan rutin, dan evaluasi berkala untuk menjaga akurasi dan relevansi.
===================================================================================================================================
Effective asset management of power transformers requires detailed data and sophisticated analysis to estimate their condition and remaining service life. This research aims to develop fault prediction and lifetime models using Dissolved Gas Analysis (DGA). The study employs four methods, the Doernenburg Ratio Method (DRM), Roger's Ratio Method (RRM), IEC Ratio Method (IRM), and Duval Triangle Method (DTM), assessing the degree of polymerization of paper insulation with furan values.
The research concludes that Random Forest algorithms perform best in predicting transformer conditions in both fault and lifetime models. Key correlations in the lifetime model include CO2 with furan (0.32), Total Combustible Gas (0.19), CO (0.15), CO2/CO (0.12), and acetylene (0.12). Meanwhile fault model accuracy varies: DRM (47%), RRM (78%), IRM (89%), DTM (96%), and DTA (98%). The Random Forest-based lifetime model shows a MAE of 38.41(ppb) and a MAPE of 5.35% after hyperparameter tuning.
Significance-weighted ranking using five methods (DRM, RRM, IRM, DTM, and DTA) achieves 98.2% accuracy, outperforming the four-method approach (92%). Testing data shows transformer condition distribution: special condition (42%), Level 1: Critical (11%), Level 2: High Risk (31%), Level 3: Moderate Risk (15%), Level 4: Low Risk (2%), and Level 5: Minimal Risk (0%). Transformers with IDs 49, 50, 47, 34, 33, and 32 require immediate action.
This study highlights the benefits of machine learning-based prediction models in power transformer management, suggesting future research to include more variables and historical data. Practical recommendations include integrating predictive models with digital monitoring systems, regular training, and periodic evaluations to maintain accuracy and relevance

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	trafo daya, fault, DGA, remaining usage lifetime, machine learning
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK2551 Electric transformers.
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Industrial Engineering > 26101-(S2) Master Thesis
Depositing User:	M Annas Albab Fauzi
Date Deposited:	17 Jul 2024 01:40
Last Modified:	17 Jul 2024 01:40
URI:	http://repository.its.ac.id/id/eprint/108370

Actions (login required)

View Item