Sistem Auto Diagnostic Kegagalan Mekanikal pada Mesin Rotasi Menggunakan Metode Random Forest untuk Mendukung Layanan Condition Monitoring

Bayhaqi, Ahmad (2026) Sistem Auto Diagnostic Kegagalan Mekanikal pada Mesin Rotasi Menggunakan Metode Random Forest untuk Mendukung Layanan Condition Monitoring. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040221010-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Mesin rotasi merupakan aset penting pada berbagai sektor industri yang memerlukan pemantauan kondisi secara berkelanjutan untuk mencegah terjadinya kegagalan mekanikal seperti Unbalance, Misalignment, dan Looseness. Pada praktik layanan Condition Monitoring, proses diagnosis masih bergantung pada interpretasi vibration analyst terhadap pola spektrum getaran sehingga memerlukan waktu analisis yang relatif lama dan berpotensi menghasilkan perbedaan interpretasi. Penelitian ini bertujuan mengembangkan sistem Auto Diagnostic berbasis Machine Learning yang mampu mengidentifikasi kondisi mesin secara otomatis dan real-time menggunakan data spektrum getaran. Data spektrum diperoleh dari Observer X melalui REST API, kemudian diproses melalui tahapan perhitungan Overall RMS, penentuan zona keparahan getaran, ekstraksi harmonik, normalisasi energi harmonik, serta feature engineering berbasis zona sehingga menghasilkan 21 fitur masukan. Model klasifikasi yang digunakan adalah Decision Tree dan Random Forest untuk mengidentifikasi empat kondisi mesin, yaitu Normal, Unbalance, Misalignment, dan Looseness. Pelatihan model menggunakan data historis industri, sedangkan validasi dilakukan secara cross-domain menggunakan Machinery Fault Simulator Lite (MFS-LT) melalui empat titik pengukuran getaran yang terintegrasi dengan dashboard dan sistem notifikasi otomatis. Hasil pengujian menunjukkan bahwa metode Random Forest memperoleh akurasi validasi real-time sebesar 97,75%, lebih tinggi dibandingkan Decision Tree sebesar 82,00%, serta mampu menyajikan hasil diagnosis dan mengirimkan notifikasi dengan latensi end-to-end rata-rata 3.280,804ms, sehingga masih berada di bawah batas satu interval polling sistem sebesar 5 detik. Hasil penelitian membuktikan bahwa sistem yang dikembangkan mampu melakukan diagnosis kondisi mesin secara otomatis, konsisten, dan real-time, serta memiliki kemampuan generalisasi yang baik pada data dengan karakteristik mesin berbeda sehingga berpotensi meningkatkan kecepatan, objektivitas, dan keandalan layanan Condition Monitoring di lingkungan industri.
====================================================================================================================================
Rotating machinery is a critical asset in various industrial sectors that require continuous condition monitoring to prevent mechanical failures such as imbalance, misalignment, and looseness. In condition monitoring practices, the diagnostic process still relies on a vibration analyst’s interpretation of vibration spectrum patterns, which requires a relatively long analysis time and may lead to differing interpretations. This research aims to develop a machine learning-based auto-diagnostic system capable of automatically identifying machine conditions in real time using vibration spectrum data. Spectrum data is obtained from Observer X via a REST API, then processed through the following steps: calculation of Overall RMS, determination of vibration severity zones, harmonic extraction, harmonic energy normalization, and zone-based feature engineering, resulting in 21 input features. The classification models used are Decision Tree and Random Forest to identify four machine conditions: Normal, Unbalance, Misalignment, and Looseness. Model training utilizes historical industrial data, while validation is performed cross-domain using the Machinery Fault Simulator Lite (MFS-LT) through four vibration measurement points integrated with a dashboard and an automatic notification system. Test results show that the Random Forest method achieved a real-time validation accuracy of 97.75%, higher than the Decision Tree’s 82.00%, and was able to present diagnostic results and send notifications with an average end-to-end latency of 3,280.804ms, which remains below the system’s single polling interval limit of 5 seconds. The research results demonstrate that the developed system is capable of diagnosing machine conditions automatically, consistently, and in real time, and exhibits good generalization capabilities across data with different machine characteristics, thereby having the potential to improve the speed, objectivity, and reliability of Condition Monitoring services in industrial environments.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Auto Diagnostic, Vibration Analysis, Condition Monitoring, Machine Learning, Decision Tree, Random Forest.
Subjects:	T Technology > T Technology (General) T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7870.23 Reliability. Failures
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Ahmad Bayhaqi
Date Deposited:	04 Aug 2026 04:21
Last Modified:	04 Aug 2026 04:21
URI:	http://repository.its.ac.id/id/eprint/142964

Actions (login required)

View Item