Desain Sistem Deteksi Anomali untuk Data Generator Diesel Tugboat Menggunakan Metode LSTM Autoencoder

Fatih, Maula Izziddin Al (2026) Desain Sistem Deteksi Anomali untuk Data Generator Diesel Tugboat Menggunakan Metode LSTM Autoencoder. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040221101-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Generator memegang peran penting dalam sistem kelistrikan tugboat, sehingga kondisi generator menjadi hal yang harus diperhatikan. Generator telah dilengkapi sistem proteksi berupa shutdown ketika parameter seperti frekuensi dan RPM melebihi threshold. Namun, sistem alarm yang digunakan saat ini masih bergantung pada alarm berbasis threshold. Penyimpangan pola data dari kondisi normal dapat menjadi indikasi potensi kerusakan, bahkan sebelum nilai data mencapai threshold. Penyimpangan pola dapat disadari jika pola atau tren data sudah dipelajari. Penelitian ini menggunakan metode Long Short-Term Memory (LSTM) Autoencoder untuk mempelajari pola operasi normal generator dan mendeteksi penyimpangan sebagai anomali. Pengumpulan dataset menghasilkan 16.452 dan 16.444 setelah dibersihkan. Supaya didapatkan model terbaik dilakukan pengujian rasio pembagian dataset dan hyperparameter tuning. Model dievaluasi menggunakan Mean Squared Error (MSE) dan Standard Deviation (STD), dengan nilai yang lebih kecil menunjukkan performa yang lebih baik. Hasil pengujian menunjukkan bahwa pembagian dataset terbaik adalah 70:15:15 yakni 70% dataset training, 15% dataset validasi, dan 15% untuk testing. Konfigurasi hyperparameter terbaik menggunakan fungsi aktivasi ReLU, timestep 10, epoch 100, dan batch size 32. Konfigurasi tersebut menghasilkan nilai MSE sebesar 0,00055 dan STD sebesar 0,00061. Nilai MSE pada data validasi kemudian digunakan sebagai dasar penentuan threshold deteksi anomali menggunakan persentil ke-99 (P99), dengan nilai 0,004981. Pengujian pada data fluktuasi RPM menunjukkan bahwa model mampu mendeteksi anomali dengan baik. Model juga berhasil diintegrasikan ke sistem monitoring sehingga hasil deteksi dapat ditampilkan pada dashboard sebagai informasi awal potensi gangguan generator.
====================================================================================================================================
Generators play a crucial role in tugboat electrical systems, so their condition is crucial. Generators are equipped with a protection system that shuts down when parameters such as frequency and RPM exceed a threshold. However, current alarm systems rely on threshold-based alarms. Deviations from normal data patterns can be an indication of potential damage, even before the data value reaches the threshold. Deviations can be detected if the data patterns or trends are studied. This study used the Long Short-Term Memory (LSTM) Autoencoder method to learn the generator's normal operating pattern and detect deviations as anomalies. The collected datasets yielded 16,452 and 16,444 after cleaning. To obtain the best model, dataset split ratio and hyperparameter tuning were tested. The model was evaluated using Mean Squared Error (MSE) and Standard Deviation (STD), with smaller values indicating better performance. The test results showed that the best dataset split ratio was 70:15:15, i.e., 70% training dataset, 15% validation dataset, and 15% for testing. The best hyperparameter configuration used the ReLU activation function, timestep 10, epoch 100, and batch size 32. This configuration produced an MSE value of 0.00055 and an STD of 0.00061. The MSE value in the validation data was then used as the basis for determining the anomaly detection threshold using the 99th percentile (P99), with a value of 0,004981. Testing on RPM fluctuation data showed that the model was able to detect anomalies well. The model was also successfully integrated into the monitoring system so that the detection results could be displayed on the dashboard as initial information on potential generator disruptions.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Tugboat, Generator, LSTM Autoencoder
Subjects:	Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK2797 Motor-generator sets. Cascade
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Maula Izziddin Al Fatih
Date Deposited:	05 Aug 2026 01:47
Last Modified:	05 Aug 2026 01:47
URI:	http://repository.its.ac.id/id/eprint/143551

Actions (login required)

View Item