Deteksi Potensi Kerusakan Mekanikal ,Bearing, Sparkplug, dan Piston Berdasarkan Data Sensor Pada Mesin Jenbacher JGS 620 di PLTMG PT. LEGI Menggunakan Autoencoder

Rachmansyah, Digdayana Rizki (2026) Deteksi Potensi Kerusakan Mekanikal ,Bearing, Sparkplug, dan Piston Berdasarkan Data Sensor Pada Mesin Jenbacher JGS 620 di PLTMG PT. LEGI Menggunakan Autoencoder. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040221131-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Mesin gas Jenbacher JGS 620 di PLTMG Teluk Lamong beroperasi dalam durasi panjang dan kondisi yang berubah, sehingga komponen Bearing, piston, dan Spark Plug berpotensi mengalami degradasi sebelum mencapai masa pakai yang direncanakan. Penelitian ini mengembangkan sistem deteksi potensi kerusakan berbasis Autoencoder dan Decision Layer menggunakan 25.672 baris data dari 11 parameter sensor. Tahap Preprocessing meliputi penanganan 820 Missing Value melalui interpolasi berbasis waktu dan Forward Fill, standardisasi Z-score, dan pembagian data secara kronologis. Skenario pembagian 70% training, 15% validation, dan 15% testing dipilih karena menghasilkan gap Train Loss dan Validation Loss terkecil sebesar 0,0097. Konfigurasi terbaik menggunakan arsitektur 11-8-6-4-6-8-11, fungsi aktivasi ELU, 200 Epoch, dan Batch Size 64 dengan MSE sebesar 0,574065. Threshold Anomali ditentukan menggunakan mean + 3σ dan menghasilkan nilai 1,3448. Pada data normal, 3.832 dari 3.850 sampel atau 99,5% berhasil diklasifikasikan sebagai kondisi normal. Pengujian Decision Layer menghasilkan precision sebesar 99,9%, 99,7%, dan 99,0%; recall sebesar 43,2%, 45,4%, dan 47,0%; serta F1-score sebesar 60,3%, 62,4%, dan 63,8% masing-masing untuk Bearing, piston, dan Spark Plug. Nilai recall dipengaruhi oleh susunan setiap skenario pengujian yang memuat tujuh hari kondisi normal dan tujuh hari pola degradasi. Decision Matrix menunjukkan kontribusi dominan oil pressure sebesar 72,25 dan knock_07 sebesar 46,34 pada Bearing; oil pressure sebesar 152,98, knock_04 sebesar 132,92, dan cyl_temp_04 sebesar 82,75 pada piston; serta cyl_temp_13 sebesar 117,90 dan knock_13 sebesar 72,18 pada Spark Plug.
=====================================================================================================================================
The Jenbacher JGS 620 gas engine at PLTMG Teluk Lamong operates for extended periods under changing conditions, so the Bearing, piston, and Spark Plug may degrade before reaching their planned service lives. This study developed an Autoencoder- and decision-layer-based potential damage detection system using 25,672 data rows from 11 sensor parameters. Preprocessing included handling 820 Missing Values through time-based interpolation and Forward Fill, Z-score standardization, and chronological data splitting. A 70% training, 15% validation, and 15% testing split was selected because it produced the smallest train-Validation Loss gap of 0.0097. The best configuration used an 11-8-6-4-6-8-11 architecture, ELU activation, 200 Epochs, and a Batch Size of 64, resulting in an MSE of 0.574065. The anomaly Threshold was determined using mean + 3σ and yielded a value of 1.3448. On normal data, 3,832 of 3,850 samples, or 99.5%, were classified as normal. Decision-layer testing produced precision values of 99.9%, 99.7%, and 99.0%; recall values of 43.2%, 45.4%, and 47.0%; and F1-scores of 60.3%, 62.4%, and 63.8% for the Bearing, piston, and Spark Plug, respectively. The recall values were influenced by the composition of each test scenario, which contained seven days of normal operation and seven days of degradation patterns. The Decision Matrix identified dominant contributions from oil pressure at 72.25 and knock_07 at 46.34 for the Bearing; oil pressure at 152.98, knock_04 at 132.92, and cyl_temp_04 at 82.75 for the piston; and cyl_temp_13 at 117.90 and knock_13 at 72.18 for the Spark Plug.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	deteksi anomali, Autoencoder, Decision Layer, Reconstruction Error, sensor PLTMG, anomaly detection, Autoencoder, Decision Layer, Reconstruction Error, PLTMG sensors.
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Digdayana Rizki Rachmansyah
Date Deposited:	06 Aug 2026 07:08
Last Modified:	06 Aug 2026 07:08
URI:	http://repository.its.ac.id/id/eprint/144184

Actions (login required)

View Item