Sistem Deteksi Anomali Sebagai Indikator Degradasi Kondisi Pada Motor Induksi Penggerak Lift Menggunakan Isolation Forest

Syailendra, Farrel Parves (2026) Sistem Deteksi Anomali Sebagai Indikator Degradasi Kondisi Pada Motor Induksi Penggerak Lift Menggunakan Isolation Forest. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040221097-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Perawatan reaktif pada motor induksi penggerak lift berisiko memicu downtime operasional akibat kegagalan deteksi gejala awal kerusakan. Penelitian ini mengusulkan sistem deteksi anomali berbasis Internet of Things pada perangkat edge computing menggunakan algoritma Isolation Forest untuk memantau getaran, arus, dan suhu secara real-time. Setiap anomali yang terdeteksi diinterpretasikan oleh metode SHAP (SHapley Additive exPlanations) untuk mengidentifikasi parameter pemicu degradasi sebagai dasar rekomendasi tindakan teknisi. Pada simulasi degradasi, Isolation Forest terbukti stabil dengan akurasi di atas 87% dan sensitivitas tinggi (arus 100%, getaran 90,7%, suhu 81,0%), jauh mengungguli Local Outlier Factor yang anjlok pada deteksi suhu (38,6%) dan getaran (59,3%). Meskipun model pembanding One-Class SVM memiliki sensitivitas sedikit lebih tinggi, Isolation Forest unggul pada efisiensi komputasi. Waktu inferensinya hanya 0,0045 milidetik per siklus, lebih ringan dari LOF yang membutuhkan waktu 0,1320 milidetik dan sekitar 100 kali lebih cepat dari One-Class SVM yang membutuhkan waktu 0,4748 milidetik. Hasil ini menyimpulkan bahwa Isolation Forest memberikan keseimbangan terbaik antara keandalan deteksi dan ringannya beban komputasi, sehingga sangat ideal untuk implementasi monitoring real-time.
=========================================================================================================================================
Reactive maintenance of elevator induction motors risks causing operational downtime due to the failure to detect early signs of damage. This study proposes an Internet of Things (IoT)-based anomaly detection system deployed on an edge computing device, utilizing the Isolation Forest algorithm to monitor vibration, current, and temperature in real-time. Each detected anomaly is interpreted using the SHAP (SHapley Additive exPlanations) method to identify the parameters triggering the degradation, serving as a basis for technician maintenance recommendations. During degradation simulations, Isolation Forest proved to be highly stable with an accuracy above 87% and high sensitivity (current 100%, vibration 90.7%, temperature 81.0%), significantly outperforming the Local Outlier Factor, which plummeted in detecting temperature (38.6%) and vibration (59.3%) anomalies. Although the comparative One-Class SVM model exhibited slightly higher sensitivity, Isolation Forest vastly excelled in computational efficiency. Its inference time was merely 0.0045 milliseconds per cycle, notably lighter than LOF, which required 0.1320 milliseconds, and approximately 100 times faster than One-Class SVM, which required 0.4748 milliseconds. These results conclude that Isolation Forest offers the optimal balance between detection reliability and computational efficiency, making it highly ideal for real-time monitoring implementations.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Motor Induksi, Predictive Maintenance, Isolation Forest, Internet of Things, Deteksi Anomali. Induction Motor, Predictive Maintenance, Isolation Forest, Internet of Things, Anomaly Detection.
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK2785 Electric motors, Induction. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Farrel Parves Syailendra
Date Deposited:	06 Aug 2026 05:59
Last Modified:	06 Aug 2026 05:59
URI:	http://repository.its.ac.id/id/eprint/144152

Actions (login required)

View Item