Nasution, Muhajirin (2024) Deteksi Anomali Pada Data Sensor Dari Kompresor CO2 Menggunakan Metode LSTM Autoencoder. Other thesis, Institut Teknologi Sepuluh Nopember.
Text
2040201022-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (6MB) | Request a copy |
Abstract
PT Pupuk Sriwidjaja merupakan pabrik pupuk yang berperan besar dalam memenuhi kebutuhan pupuk dan bahan kimia pertanian di Indonesia. Perusahaan ini memproduksi pupuk utama seperti urea, amonia, dan NPK, serta produk samping seperti CO2 cair, CO2 padat, oksigen, dan nitrogen. Proses produksi urea melibatkan enam unit, dimana setiap unit memerlukan kompresor CO2 untuk menunjang peningkatan tekanan operasional. Hal ini membuat kompresor CO2 menjadi elemen kritis dalam proses produksi, khususnya pada sintesa unit, dengan tekanan operasional yang dibutuhkan adalah 175 kg/cm². Oleh karena itu, kerusakan atau kegagalan kompresor dapat berdampak fatal karena setiap unit proses membutuhkan tekanan yang dihasilkan oleh kompresor CO2, khususnya pada sintesa unit. Pada sistem yang sedang berjalan digunakan ambang batas statis dalam mendeteksi anomali pada sensor pressure discharge, temperature discharge, dan flow discharge yang berada pada kompresor. Ambang batas statis ini memiliki kelemahan karena tidak dapat mendeteksi anomali pada data temporal yang memiliki perubahan drastis. Untuk mengatasi permasalahan tersebut dikembangkanlah metode pendeteksian anomali menggunakan LSTM Autoencoder dengan pada sensor pressure discharge, temperature discharge dan flow discharge kompresor CO2. Adapun hasil dari penelitian, model LSTM Autoencoder mampu mendeteksi anomali pada data testing dengan menggunakan threshold dari fase validation, didapatkan 61 datapoint anomali dari setiap sensor dan dari hasil perbandingan model yang dilakukan sebanyak lima kali percobaan dengan merubahan parameter timesteps, dimulai dari panjang timesteps 5, 10, 15, 20 dan 30. Didapatkan model realtif baik dengan MSE (Mean Squared Error) terkecil pada nilai 0,0074 pada panjang lima timesteps.
=====================================================================================================================================
PT Pupuk Sriwidjaja is a fertilizer plant that plays a significant role in meeting the needs for fertilizers and agricultural chemicals in Indonesia. The company produces primary fertilizers such as urea, ammonia, and NPK, as well as by-products such as liquid CO2, solid CO2, oxygen, and nitrogen. The urea production process involves six units, each of which requires a CO2 compressor to support the increase in operational pressure. This makes the CO2 compressor a critical element in the production process, particularly in the synthesis unit, with the required operational pressure being 175 kg/cm². Therefore, damage or failure of the compressor can have fatal impacts as each process unit requires the pressure generated by the CO2 compressor, especially in the synthesis unit. In the current system, a static threshold is used to detect anomalies in the pressure discharge, temperature discharge, and flow discharge sensors located on the compressor. This static threshold has a weakness as it cannot detect anomalies in temporal data that has drastic changes. To address this issue, an anomaly detection method was developed using LSTM Autoencoder on the pressure discharge, temperature discharge, and flow discharge sensors of the CO2 compressor. The research results showed that the LSTM Autoencoder model was able to detect anomalies in the testing data using the threshold from the validation phase. A total of 61 anomaly data points were detected from each sensor, and from the comparison of models carried out in five trials with changes in the timesteps parameter, starting from timestep lengths of 5, 10, 15, 20, and 30. A relatively good model was obtained with the smallest Mean Squared Error (MSE) value of 0.0074 at a timestep length of five.
Item Type: | Thesis (Other) |
---|---|
Uncontrolled Keywords: | anomali, autoencoder, kompresor, LSTM, sensor |
Subjects: | T Technology > T Technology (General) > T174 Technological forecasting |
Divisions: | Faculty of Vocational > 36304-Automation Electronic Engineering |
Depositing User: | Muhajirin Nasution |
Date Deposited: | 23 Aug 2024 02:22 |
Last Modified: | 23 Aug 2024 02:22 |
URI: | http://repository.its.ac.id/id/eprint/115507 |
Actions (login required)
View Item |