Sistem Prediksi Kondisi Transformator Electric Furnace Berbasis Data Dissolved Gas Analysis Menggunakan Machine Learning

Ramadhan, Muhammad Fahmi (2025) Sistem Prediksi Kondisi Transformator Electric Furnace Berbasis Data Dissolved Gas Analysis Menggunakan Machine Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5009211007-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Transformator electric arc furnace (EAF) merupakan aset krusial dalam industri peleburan nikel, beroperasi di bawah beban dinamis ekstrem yang rentan memicu kegagalan sehingga diagnosis kondisi transformator melalui Dissolved Gas Analysis (DGA) menjadi penting, sehingga dikembangkan sistem prediksi kondisi transformator electric furnace. Metodologi penelitian ini mencakup pra-pemrosesan data, diikuti oleh rekayasa fitur berbasis statistik bergulir (rolling statistics). Support Vector Regression (SVR) dipilih karena keandalannya pada dataset berukuran kecil, dengan optimisasi hyperparameter dilakukan melalui GridSearchCV dan validasi silang TimeSeriesSplit untuk menjaga dependensi temporal. Hasil pengujian menunjukkan akurasi diagnosis kondisi keseluruhan yang menjanjikan. Analisis pada gas-gas kunci menunjukkan model memiliki performa yang relatif kurang baik dalam memprediksi gas etilena (C₂H₄) tetapi cukup baik dalam memprediksi gas asetilena (C₂H₂). Pada data uji, performa prediksi gas etilena tercatat dengan RMSE 9,354 ppm, MAE 7,243 ppm, dan SMAPE 29,811% sedangkan performa prediksi gas asetilena tercatat dengan RMSE 4,37 ppm, MAE 2,775 ppm, dan SMAPE 14,303%. Pada skenario peramalan delapan periode mendatang, performa prediksi meningkat dengan RMSE 2,11 ppm, MAE 1,82 ppm, dan SMAPE 12,54% untuk gas asetilena dan RMSE 3,48 ppm, MAE 2,91 ppm, dan SMAPE 9,89%. Meskipun terdapat kesulitan pada prediksi data uji gas etilena, sistem ini tetap menunjukkan potensi yang signifikan. Pada skenario peramalan, diagnosis kondisi akhir transformator berdasarkan metode Total Dissolved Combustible Gases (TDCG) berhasil mencapai akurasi 100%, menunjukkan kemampuannya mengidentifikasi tren degradasi secara keseluruhan.
=====================================================================================================================================
Electric arc furnace (EAF) transformers are crucial assets in the nickel smelting industry, operating under extreme dynamic loads that are prone to failure, making it important to diagnose the condition of transformers through Dissolved Gas Analysis (DGA). Therefore, a system for predicting the condition of electric furnace transformers was developed. The research methodology includes data pre-processing, then feature engineering using rolling statistics-based feature. Support Vector Regression (SVR) was chosen for its reliability on small datasets, with hyperparameter optimization performed via GridSearchCV and TimeSeriesSplit cross-validation to maintain temporal dependencies. Testing results show promising overall condition diagnosis accuracy. Analysis of key gases indicates that the model performs relatively poorly in predicting ethylene gas (C₂H₄) but adequately in predicting acetylene gas (C₂H₂). On the test data, the prediction performance for ethylene gas was recorded with an RMSE of 9.354 ppm, MAE of 7.243 ppm, and SMAPE of 29.811%, while the prediction performance for acetylene gas was recorded with an RMSE of 4.37 ppm, MAE of 2.775 ppm, and SMAPE of 14.303%. In the eight-period forecasting scenario, prediction performance improved with the RMSE of 2.11 ppm, MAE of 1.82 ppm, and SMAPE of 12.54% for acetylene gas, and an RMSE of 3.48 ppm, MAE of 2.91 ppm, and SMAPE of 9.89%. Despite the challenges in predicting ethylene gas test data, the system still demonstrates significant potential. In the forecasting scenario, the final transformer condition diagnosis based on the Total Dissolved Combustible Gases (TDCG) method successfully achieved 100% accuracy, showcasing its capability to identify overall degradation trends.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Dissolved Gas Analysis, Diagnosis Kondisi, Support Vector Regression, Prediksi Deret Waktu, Transformator Electric Furnace, Dissolved Gas Analysis (DGA), Electric Furnace Transformer, Fault Diagnosis Support Vector Regression (SVR), Time Series Forecasting
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK2551 Electric transformers. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK6565.T7 Transformers
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Physics Engineering > 30201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Fahmi Ramadhan
Date Deposited:	06 Aug 2025 09:33
Last Modified:	06 Aug 2025 09:33
URI:	http://repository.its.ac.id/id/eprint/127695

Actions (login required)

View Item