Sistem Deteksi Anomali Tekanan Pada Jaringan Kompresor Udara Menggunakan Single Board Computer Dengan Metode Isolation Forest

Gunawan, Theo Andre (2026) Sistem Deteksi Anomali Tekanan Pada Jaringan Kompresor Udara Menggunakan Single Board Computer Dengan Metode Isolation Forest. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040221039-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Sistem distribusi udara terkompresi merupakan utilitas kritis pada industri manufaktur karena kestabilan tekanan pada receiver tank berperan penting dalam menjaga kontinuitas proses produksi. Namun, sistem monitoring tekanan di PT Kerta Rajasa Raya Cabang Mojosari masih dilakukan secara manual sehingga data tekanan tidak terekam secara kontinu, tidak dapat dipantau secara real-time, serta belum tersedia data historis sebagai dasar evaluasi kondisi operasi. Selain itu, operator harus melakukan inspeksi langsung pada setiap receiver tank sehingga penyimpangan pola operasi berpotensi terlambat teridentifikasi. Padahal, perubahan pola tekanan merupakan indikator awal gangguan pada sistem distribusi udara terkompresi sebelum berkembang menjadi kondisi under pressure maupun over pressure. Oleh karena itu, penelitian ini bertujuan mengembangkan sistem monitoring tekanan berbasis Internet of Things (IoT) yang mampu melakukan akuisisi, visualisasi data secara real-time, serta deteksi anomali sebagai mekanisme early warning. Sistem dikembangkan menggunakan Raspberry Pi sebagai single-board computer, sedangkan deteksi anomali dilakukan menggunakan metode Isolation Forest terhadap komponen seasonal hasil dekomposisi Seasonal-Trend Decomposition using Loess (STL). Kinerja metode dievaluasi menggunakan enam skenario Forward Validation dan Backward Validation serta dibandingkan dengan metode One-Class Support Vector Machine (OCSVM). Hasil penelitian menunjukkan bahwa Isolation Forest menghasilkan performa deteksi yang lebih konsisten pada seluruh skenario pengujian dengan anomaly rate sebesar 3,01%–3,09%, sedangkan OCSVM menghasilkan anomaly rate sebesar 2,67%–3,23%. Berdasarkan hasil tersebut, Isolation Forest dipilih sebagai metode deteksi terbaik karena memberikan hasil deteksi yang lebih stabil terhadap pola operasi normal sistem. Dengan demikian, sistem yang dikembangkan mampu mendukung monitoring tekanan secara real-time serta menyediakan mekanisme early warning untuk identifikasi dini terhadap potensi gangguan pada sistem distribusi udara terkompresi.
======================================================================================================================================
Compressed air distribution systems are critical utilities in manufacturing industries, where maintaining stable pressure in receiver tanks is essential to ensure production continuity. However, pressure monitoring at PT Kerta Rajasa Raya Mojosari Branch is still performed manually, resulting in discontinuous data acquisition, the absence of real-time monitoring, and limited historical data for operational evaluation. In addition, operators are required to conduct on-site inspections at each receiver tank, causing deviations in operating pressure patterns to be identified late. In practice, changes in pressure patterns represent an early indication of disturbances in compressed air distribution systems before developing into under-pressure or over-pressure conditions. Therefore, this study proposes an Internet of Things (IoT)-based pressure monitoring system capable of real-time data acquisition, visualization, and anomaly detection as an early warning mechanism. The system employs a Raspberry Pi as a single-board computer, while anomaly detection is performed using the Isolation Forest algorithm on the seasonal component extracted through Seasonal-Trend Decomposition using Loess (STL). The proposed method was evaluated using six Forward Validation and Backward Validation scenarios and compared with the One-Class Support Vector Machine (OCSVM) algorithm. Experimental results demonstrate that Isolation Forest provides more consistent detection performance across all evaluation scenarios, achieving an anomaly rate of 3.01% to 3.09%, whereas OCSVM produces an anomaly rate ranging from 2.67% to 3.23%. Based on these findings, Isolation Forest was selected as the most suitable anomaly detection method due to its more stable detection performance in representing normal operating patterns. Consequently, the developed system enables real-time pressure monitoring while providing an effective early warning mechanism for the early identification of potential disturbances in compressed air distribution systems.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deteksi Anomali, Internet of Things (IoT), Isolation Forest, Monitoring Tekanan, Receiver Tank.
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7878 Electronic instruments
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Theo Andre Gunawan
Date Deposited:	05 Aug 2026 04:20
Last Modified:	05 Aug 2026 04:20
URI:	http://repository.its.ac.id/id/eprint/144033

Actions (login required)

View Item