Rahmadi, Rahmadi (2026) Deteksi Dini Kegagalan Rolling Element Bearing Pada Rotating Equipment PLTGU Grati Menggunakan Envelope Spectrum Analysis Dengan Validasi ROC/AUC Dan Penetapan Site-Specific Threshold. Masters thesis, Istitut Teknologi Sepuluh Nopember.
![[thumbnail of 6007241021-Master_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
6007241021-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy |
Abstract
Keandalan rotating equipment pada PLTGU Grati sangat menentukan kontinuitas produksi listrik. Dalam praktik pemantauan kondisi, waktu peringatan dini (lead time) yang dibutuhkan untuk perencanaan pemeliharaan sering tidak memadai, terutama ketika evaluasi hanya mengandalkan parameter overall vibration (velocity RMS) berbasis ISO 10816-3 yang rentan mengalami masking terhadap sinyal cacat awal bearing. Penelitian ini bertujuan merumuskan standar ambang batas (site-specific threshold) operasional untuk deteksi dini kegagalan rolling element bearing pada rotating equipment PLTGU Grati menggunakan Envelope Spectrum Analysis, yang divalidasi secara objektif melalui pendekatan statistik ROC/AUC serta verifikasi fisik, guna menjamin lead time pemeliharaan yang memadai.
Pendekatan yang digunakan adalah studi kasus komparatif-kuantitatif retrospektif berbasis data run-to-failure pada lima peralatan kritis: Motor Condensate Extraction Pump (CEP), Torque Converter output shaft (TC2.3), Motor Circulating Water Pump (CWP), Motor Auxiliary Oil Pump (AOP), dan Pompa Closed Cycle Cooling Water Pump (C3WP). Data overall RMS dan time waveform diproses menjadi spektrum envelope untuk mengidentifikasi frekuensi cacat (BPFI/BPFO/BSF/FTF) dan tren amplitudo. Untuk menyetarakan perbedaan karakteristik tiap mesin, fitur envelope dinormalisasi terhadap baseline kondisi sehat. Validasi diagnosis dilakukan melalui verifikasi fisik (teardown report) serta evaluasi performa klasifikasi menggunakan kurva ROC/AUC. Optimasi titik ambang dilakukan dengan Youden’s Index untuk memperoleh threshold alarm yang paling efektif.
Hasil menunjukkan Envelope Spectrum Analysis secara konsisten lebih sensitif dibanding overall RMS, menghasilkan rata-rata lead time 79 hari pada lima kasus, dengan rata-rata amplifikasi sinyal 15,7x terhadap baseline. Evaluasi ROC menunjukkan performa klasifikasi yang sangat tinggi untuk envelope (AUC ≈ 0,99) dibanding overall RMS (AUC ≈ 0,78). Berdasarkan optimasi Youden, direkomendasikan threshold peringatan 2x baseline untuk mesin berdinamika stabil dan 3x baseline untuk torque converter, serta threshold alarm universal ≥4x baseline. Implementasi threshold site-specific ini berpotensi meningkatkan keandalan serta menekan risiko forced outage dan biaya downtime secara signifikan.
======================================================================================================================================
The reliability of rotating equipment at PLTGU Grati is critical to ensure continuous electricity generation. In practical condition monitoring, the early-warning time (lead time) required for maintenance planning is often insufficient, particularly when evaluation relies solely on overall vibration (velocity RMS) in accordance with ISO 10816-3, which is susceptible to masking of weak early-stage bearing fault signatures. This study aims to formulate an operational site-specific threshold standard for the early detection of rolling element bearing failures in rotating equipment at PLTGU Grati using Envelope Spectrum Analysis, which is objectively validated through ROC/AUC statistical approaches and physical verification, to ensure a sufficient maintenance lead time.
A retrospective, comparative–quantitative case study approach was applied using industrial run-to-failure data from five critical assets: Condensate Extraction Pump (CEP) motor, Torque Converter output shaft (TC2.3), Circulating Water Pump (CWP) motor, Auxiliary Oil Pump (AOP) motor, and Closed Cycle Cooling Water Pump (C3WP). Overall RMS and acceleration time waveforms were processed into envelope spectra to identify defect frequencies (BPFI/BPFO/BSF/FTF) and to track amplitude trends. To ensure comparability across different machines, envelope features were normalized to each machine’s healthy baseline. Diagnostic interpretation was validated through physical verification (teardown reports) and classification performance was quantified using ROC/AUC analysis. Threshold optimization was performed using Youden’s Index to obtain the most effective alarm limit.
Results indicate that Envelope Spectrum Analysis is consistently more sensitive than overall RMS, providing an average lead time of 79 days across the five cases, with a mean signal amplification of 15.7× relative to baseline. ROC evaluation shows excellent classification performance for the envelope feature (AUC ≈ 0.99) compared with overall RMS (AUC ≈ 0.78). Based on Youden optimization, a warning threshold of 2× baseline is recommended for stable-dynamics machines and 3× baseline for the torque converter, while a universal alarm threshold of ≥4× baseline is proposed. Implementing these site-specific thresholds has the potential to enhance reliability and significantly reduce the risk of forced outages and downtime costs.
Actions (login required)
 |
View Item |