Pengembangan Model Predictive Maintenance Pada Mesin Main Motor Raw Mill

Zakiyyah, Fauziyyah Firdausi (2026) Pengembangan Model Predictive Maintenance Pada Mesin Main Motor Raw Mill. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6010232007-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Downtime akibat kegagalan mesin yang terjadi secara tiba-tiba dapat menyebabkan kerugian yang signifikan bagi perusahaan industri. Oleh karena itu, diperlukan strategi perawatan yang mampu mendeteksi potensi degradasi mesin sejak dini guna meminimalkan risiko kegagalan. Pemantauan kondisi mesin industri dapat dilakukan melalui pemanfaatan data sensor kondisi yang dihasilkan selama proses operasi. Namun, pengelolaan data sensor tersebut memerlukan penerapan teknik analitik yang andal agar dapat memberikan informasi yang bernilai dalam mendukung implementasi predictive maintenance. Penelitian ini bertujuan untuk mengembangkan model predictive maintenance berbasis machine learning melalui integrasi model diagnosis dan model prognosis pada mesin main motor raw mill. Penelitian dilakukan melalui beberapa tahapan utama, yaitu pengembangan model diagnosis kondisi mesin, pengembangan model prognosis berbasis Remaining Useful Life (RUL), serta evaluasi kinerja model. Model diagnosis dikembangkan menggunakan data multi sensor untuk mengklasifikasikan kondisi kesehatan mesin ke dalam beberapa kelas kondisi operasional. Selanjutnya, model prognosis dikembangkan dengan memanfaatkan konsep RUL yang diformulasikan sebagai klasifikasi fase degradasi mesin menggunakan algoritma Long Short-Term Memory (LSTM) guna mengidentifikasi kondisi early fault. Pemisahan data dilakukan secara time-based untuk menghindari terjadinya data leakage, serta diterapkan hyperparameter tuning untuk memperoleh konfigurasi model yang terbaik. Hasil pengujian menunjukkan bahwa pada tahap diagnosis, model dengan algoritma XGBoost menghasilkan kinerja terbaik dengan tingkat akurasi 98,8%. Sementara itu, pada tahap prognosis model LSTM dengan parameter terpilih memberikan kinerja klasifikasi RUL yang baik, yang ditunjukkan oleh nilai Area Under Curve (AUC) sebesar 0,944 pada pengujian data baru. Hal tersebut menunjukkan kemampuan model dalam membedakan kondisi mesin pada fase normal dan early fault. Dengan demikian, model yang dikembangkan dapat digunakan sebagai pendukung pengambilan keputusan pemeliharaan dalam implementasi predictive maintenance.
==================================================================================================================================
Downtime due to sudden machine failures can cause significant losses for industrial companies. Therefore, a maintenance strategy capable of detecting potential machine degradation at an early stage is required to minimize the risk of failure. Machine condition monitoring can be conducted through the utilization of condition sensor data generated during the operation process. However, the processing of sensor data requires the application of reliable analytical techniques to provide meaningful information that supports the implementation of predictive maintenance. This study was conducted to develop a machine learning–based predictive maintenance model through the integration of diagnostic and prognostic models applied to a raw mill main motor. The research was carried out through several main stages, including the development of a machine condition diagnosis model, the development of a prognosis model based on Remaining Useful Life (RUL), and the evaluation of model performance. The diagnostic model was developed using multi-sensor data to classify machine health conditions into several operational condition classes. Furthermore, the prognostic model was developed by utilizing the RUL concept formulated as a classification of machine degradation phases using the Long Short-Term Memory (LSTM) algorithm to identify early fault conditions. Data splitting was performed using a time-based approach to avoid data leakage, and hyperparameter tuning was applied to obtain the optimal model configuration. The experimental results showed that, at the diagnosis stage, the XGBoost-based model achieved the best performance with an accuracy of 98.8%. Meanwhile, at the prognosis stage, the LSTM model with selected parameters demonstrated good RUL classification performance, as indicated by an Area Under the Curve (AUC) value of 0.944 on unseen test data. These results indicate that the proposed model is capable of distinguishing machine conditions between normal and early fault phases. Therefore, the developed model can be utilized as a decision support tool for maintenance planning in the implementation of predictive maintenance.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Downtime, XGBoost, LSTM, Machine Learning, Predictive Maintenance, Remaining Useful Life (RUL)
Subjects:	T Technology > TA Engineering (General). Civil engineering (General) > TA169 Reliability (Engineering)
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Industrial Engineering > 26101-(S2) Master Thesis
Depositing User:	Fauziyyah Firdausi Zakiyyah
Date Deposited:	22 Jan 2026 02:04
Last Modified:	22 Jan 2026 02:04
URI:	http://repository.its.ac.id/id/eprint/130008

Actions (login required)

View Item