Prediksi Remaining Useful Life (RUL) Pada Pompa Booster Dengan Metode Similarity Measures

Widyantoro, Galih (2023) Prediksi Remaining Useful Life (RUL) Pada Pompa Booster Dengan Metode Similarity Measures. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 10311910000052-Undergraduate_Thesis for D4 program.pdf - Accepted Version Restricted to Repository staff only until 1 October 2025. Download (5MB) | Request a copy

Abstract

Pengolahan bijih tambang PT Freeport Indonesia (PTFI) dilakukan di Pabrik Concentrating dengan hasil akhir berupa bubur konsentrat (slurry) yang mengandung mineral berharga (tembaga, emas dan perak). Slurry ditransfer oleh pompa dari Unit Pump House menuju Pelabuhan Amamapare sejauh 110 Km, sehingga unit ini memiliki peran penting dalam proses produksi. Pada Unit Pump House terdapat pompa booster. Salah satu permasalahan pada pompa booster adalah sistem preventive maintenance (PM) yang direncanakan terjadi sebelum jadwal, sehingga jadwal maintenance yang sudah ada berganti menjadi tidak menentu. Hal tersebut mendasari penelitian ini untuk membuat sistem predictive maintenance (PdM) sebagai rekomendasi penjadwalan maintenance berdasarkan kondisi aktual pompa. Sistem PdM memanfaatkan Machine Learning dengan metode Similarity Measures untuk memprediksi Remaining Useful Life (RUL) yakni sisa umur operasi dari pompa booster. Pemodelan Health Index (HI) dibuat menggunakan beberapa variabel riwayat data instrumen pompa booster. Pemilihan variabel dilakukan dengan melihat trendabillity, kemudian terpilihlah arus motor pompa, tekanan pada outlet dan durasi penggunaan optimum (run hour). Pemodelan menggunakan model regresi linear berganda dan distabilkan dengan regresi polinomial orde 2. Setelah HI didapatkan, Similarity Measures yang dilakukan menggunakan metode Euclidean Distance (ED). Dari pengukuran ED, HI data uji yang memiliki jarak terkecil dengan data latih digunakan sebagai prediksi RUL. Sistem yang dirancang berhasil untuk melakukan prediksi RUL pompa booster berdasarkan data latihnya. Variabel run hour memiliki pengaruh 1,0533 berkorelasi negatif, arus motor pompa 0,006 berkorelasi positif dan tekanan outlet 0,3093 berkorelasi positif terhadap nilai HI. Ketiga variabel tersebut mendapatkan nilai r-square 0,78 yang berarti dapat menjelaskan HI dari pompa sebesar nilai tersebut dan masuk dalam kategori memiliki pengaruh kuat. Pengukuran similarity measures dengan Euclidean berhasil memprediksi kemiripan pola lintasan degradasi HI dengan nilai mean absolute percentage error (MAPE) 7,9% yang masuk dalam kategori memiliki kemampuan prediksi sangat baik.
==================================================================================================================================
PT Freeport Indonesia (PTFI) ore is processed into a concentrate slurry containing valuable minerals (copper, gold and silver). The Pump House Unit plays an important role in the production process as the slurry is pumped 110 km to the Amamapare Port. There is also a booster pump located in the Pump House Unit. One of the issues with the booster pump is the Preventive Maintenance (PM) system that is scheduled ahead of schedule, making the existing maintenance schedule irregular. This underlies this research to create a Predictive Maintenance (PdM) system as a maintenance scheduling recommendation based on the pump's actual condition. The PdM system uses machine learning with similarity measures to predict Remaining Useful Life (RUL), which is the remaining life of the booster pump. The Health Index (HI) modeling is performed using multiple variables from the booster pump instrument data history. The variables are selected based on trendability. The selected variables are pump motor current, outlet pressure, and optimum operating hours. Modeling uses multiple linear regression models stabilized by 2nd order polynomial regression. After the HI was obtained, similarity measures were performed using the Euclidean Distance (ED) method. The RULs are predicted from the HIs of the test samples with the least distance from the training samples. The designed system is successful in predicting the RUL of the booster pump based on the training data. Running hour variable has 1.0533 negative correlation, pump motor current 0.006 positive correlation, and outlet pressure 0.3093 positive correlation with HI value. The three variables have an r-squared value of 0.78, meaning it can explain the HI of the pump. Similarity measures with Euclidean successfully predicted the similarity of HI degradation trajectory patterns with Mean Absolute Percentage Error (MAPE) value of 7.9%. This falls into the category of having very good predictive ability

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Euclidean Distance, Machine Learning, Pompa Slurry Sentrifugal, Remaining Useful Life, Similarity Measures, Centrifugal Slurry Pump
Subjects:	H Social Sciences > HD Industries. Land use. Labor > HD56.25 Industrial efficiency--Measurement. Industrial productivity--Measurement.
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Galih Widyantoro
Date Deposited:	08 Aug 2023 00:59
Last Modified:	08 Aug 2023 00:59
URI:	http://repository.its.ac.id/id/eprint/103989

Actions (login required)

View Item