Model Prediksi Diagnosis Dan Pronosis Multi Failure Mode Pada Peralatan High Pressure Feedwater Heater PLTU

Fano, Naufal Firjatulloh (2026) Model Prediksi Diagnosis Dan Pronosis Multi Failure Mode Pada Peralatan High Pressure Feedwater Heater PLTU. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6026242015-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

PLTU merupakan produsen energi listrik terbesar di Indonesia, sehingga penting untuk menjaga efisiensi dan keandalan operasional PLTU agar pasokan listrik tidak terganggu. Salah satu peralatan krusial dalam optimalisasi siklus Rankine pada pembangkit listrik berbasis termal adalah High Pressure (HP) Feedwater Heater, yang berfungsi menaikkan suhu air umpan dengan memanfaatkan sisa uap panas turbin untuk meminimalisir penggunaan bahan bakar dalam produksi uap bertekanan tinggi. Sebagai upaya menjaga keandalan peralatan, dilakukan strategi predictive maintenance yang bersifat proaktif untuk meminimalisir biaya perbaikan dan unplanned downtime, yang kini didukung oleh pendekatan AI dan ML berdasarkan data histori peralatan. Penelitian ini melakukan perancangan data degradasi dan model prediksi prognosis multi mode kegagalan pada peralatan HP Feedwater Heater PLTU. Data degradasi dirancang melalui proses fusion antara aspek data-driven dan knowledge-informed, yang digabungkan dengan mekanisme adaptive weighting untuk menghasilkan nilai komposit Failure Score pada setiap mode kegagalan. Model prediksi prognosis dirancang untuk melakukan prediksi kemungkinan terjadinya failure mode event pada beberapa periode ke depan dengan strategi direct multistep forecasting dan pembobotan spike-aware, menggunakan algoritma ensemble learning dan deep learning. Hasil validasi menunjukkan bahwa nilai Failure Score mampu merepresentasikan kondisi degradasi peralatan secara konsisten pada ketiga mode kegagalan, dibuktikan melalui perbandingan dengan corrective maintenance log pada mode Tube Leakage yang berhasil mendeteksi 3 dari 5 kejadian pemeliharaan secara dini dengan lead time tertinggi mencapai 97,58 jam. Mekanisme adaptive weighting juga terbukti lebih unggul dibandingkan static weighting dalam mendeteksi pola degradasi halus yang tidak teridentifikasi oleh aturan pakar. Pada tahap prediksi prognosis, algoritma CatBoost menghasilkan performa Recall tertinggi secara konsisten pada ketiga mode kegagalan, yaitu sebesar 83,30% pada Tube Fouling, 76,35% pada Tube Leakage, dan 98,70% pada Control Valve Stuck.
================================================================================================================================
Coal-fired power plants are the largest producers of electricity in Indonesia, so it is important to maintain their operational efficiency and reliability to ensure an uninterrupted electricity supply. One of the critical pieces of equipment in optimizing the Rankine cycle at thermal power plants is the High-Pressure (HP) Feedwater Heater, which raises the temperature of the feedwater by utilizing residual steam heat from the turbine to minimize fuel consumption in the production of high-pressure steam. To ensure equipment reliability, a predictive maintenance strategy is implemented to minimize repair costs and unplanned downtime, which can now be supported by AI and ML approaches based on historical equipment data. This study proposed a degradation data model and a multi-mode failure prognosis prediction model for HP Feedwater Heater equipment in coal-fired power plants. The degradation data was constructed through a fusion of data-driven and knowledge-informed aspects, combined with an adaptive weighting mechanism to generate a composite Failure Score for each failure mode. The prognostic prediction model was designed to predict the likelihood of failure mode events occurring over several future periods using a direct multistep forecasting strategy and spike-aware weighting, employing ensemble learning and deep learning algorithms. The validation results show that the Failure Score consistently represents equipment degradation across all three failure modes, as demonstrated by a comparison with corrective maintenance logs for the Tube Leakage mode, which successfully detected 3 out of 5 maintenance events early, with a maximum lead time of 97.58 hours. The adaptive weighting mechanism also proved superior to static weighting in detecting subtle degradation patterns not yet identified by expert rules. In the prognosis prediction stage, the CatBoost algorithm consistently achieved the highest Recall performance across all three failure modes: 83.30% for Tube Fouling, 76.35% for Tube Leakage, and 98.70% for Control Valve Stuck.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Prediksi kegagalan, Diagnosis, Prognosis, Degradasi performa, HP Feedwater Heater, Failure Prediction, Diagnosis, Prognosis, Performance Degradation, HP Feedwater Heater
Subjects:	Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 59101-(S2) Master Thesis
Depositing User:	Naufal Firjatulloh Fano
Date Deposited:	29 Jul 2026 02:10
Last Modified:	29 Jul 2026 02:10
URI:	http://repository.its.ac.id/id/eprint/139015

Actions (login required)

View Item