Penerapan Metode Random Forest untuk Deteksi Stiction pada Katup Kontrol

MAULANA ISKAK, Dr.-Ing. Awang N. I. Wardana, S.T., M.T., M.Sc., Ir. Nopriadi, S.T., M.Sc., Ph.D.

2021 | Skripsi | S1 TEKNIK FISIKA

Kalang kontrol di industri merupakan komponen yang harus selalu dijaga performanya karena akan menentukan performa dari pabrik. Osilasi pada kalang kontrol mengindikasikan bahwa kalang kontrol tersebut mempunyai performa yang buruk. Sebanyak 20-30% dari kalang kontrol yang berosilasi di industri disebabkan oleh permasalahan pada katup kontrol khususnya stiction. Stiction dapat meningkatkan variabilitas pada proses dan menyebabkan kondisi berbahaya pada pabrik sehingga perlu dideteksi sedini mungkin. Kemudian, memasuki era Industri 4.0, terjadi perubahan paradigma pada progran deteksi stiction yang awalnya berbasis aturan (rule-based) dengan mode offline menjadi berbasis machine learning dengan mode online. Pada penelitian ini, dibangun program deteksi stiction online menggunakan metode random forest. Data-data variabel keluaran controller (OP) dan data variabel proses (PV) akan disegmentasi pada ukuran jendela data tertentu dan kemudian dilakukan ekstraksi fitur time series. Hasil ekstraksi kemudian digunakan untuk membangun model random forest yang mampu melakukan tugas deteksi stiction. Model kemudian diimplementasikan secara online dengan bantuan jendela data bergeser (sliding windows) dan MQTT. Hasil penelitian menunjukkan bahwa model random forest mempunyai performa terbaik ketika ukuran jendela data sebesar 100 dengan n trees = 100, maximum features = 19, minimum samples split = 20, dan maximum depth = 70. Berdasarkan pengujian secara online yang telah dilakukan, program dapat melakukan deteksi dengan benar pada 17 dari 19 kalang kontrol SACAC.

The control loop in the industry is a component that must be maintained because it will determine the performance of the factory. Oscillations in the control loop indicate that the control loop has poor performance. About 20-30% of the oscillating control loop in the industry is caused by the stiction of the control valve. Stiction can increase the variability in the process and cause hazardous conditions in the plant so it needs to be detected as early as possible. Entering the Industrial 4.0 era, there was a paradigm shift in the stiction detection program which was originally rule-based in offline mode to machine learning-based in online mode. In this study, an online stiction detection program was built using the random forest method. The controller output variable data (OP) and process variable data (PV) will be segmented on a specified data window size. After that, time-series feature extraction is performed on the data. The extraction results are then used to build a random forest model that can detect stiction in the control valve. The model is implemented online with the help of sliding windows and MQTT. The results showed that the random forest model had the best performance when the data window size was 100 with n trees = 100, maximum features = 19, minimum samples split = 20, and maximum depth = 70. Based on the online tests that have been carried out, the program was able to correctly detect 17 of the 19 SACAC control loops.

Kata Kunci : Control Valve, Stiction, Machine Learning, Random Forest