PENERAPAN METODE KLASIFIKASI K-NEAREST NEIGHBORS (KNN) UNTUK DETEKSI STICTION PADA KATUP KONTROL

REMASITA CAHYANING PUTRI, Dr.-Ing. Awang N. I. Wardana, S.T., M.T., M.Sc.; Nopriadi, S.T., M.Sc., Ph.D.

2022 | Skripsi | S1 TEKNIK FISIKA

Stiction pada katup kontrol sering menjadi penyebab buruknya performa kalang kontrol dan dapat merugikan pabrik sehingga perlu dideteksi lebih dini. Stiction terjadi ketika stem pada katup kontrol tidak bergerak sesuai dengan sinyal keluaran kontroler. Dimulainya era industri 4.0 dan dikenalnya konsep pabrik cerdas membuat deteksi stiction yang semula dibuat dengan mode offline, kini dapat dibuat dengan mode online dengan didukung pemanfaatan machine learning. Pada penelitian ini, program deteksi stiction pada katup kontrol akan dibangun dengan menggunakan metode k-Nearest Neighbors (KNN) dan akan diimplementasikan secara online agar proses deteksi dapat berjalan dengan cepat dan tepat waktu (realtime). KNN merupakan metode supervised machine learning yang memungkinkan pengklasifikasian data ke dalam kelompok tertentu berdasarkan data yang telah dipelajari. Data variabel proses (PV) dan keluaran controller (OP) akan diproses menggunakan ekstraksi fitur time series dan penyekalaan fitur. Hasilnya digunakan untuk membangun model. Model yang didapat kemudian diimplementasikan secara online dengan menggunakan jendela bergeser dan MQTT. Dari hasil penelitian, didapatkan model KNN dengan k = 5, ukuran jendela data = 150, dan rumus jarak manhattan dan telah diimplementasikan pada program deteksi stiction online. Hasil pengujian secara online menunjukkan bahwa program dapat mendeteksi 32 dari 34 kalang kontrol SACAC dengan benar.

Stiction on the control valve is often causing a poor performance on the control loop and can harm the plant. Therefore, it needs to be detected as early as possible. Stiction occurs when the stem on the control valve does not move according to the controller's output signal. The beginning of Industrial Revolution 4.0 and the smart factory enabled stiction detection, which was originally created in offline mode, to be created in online mode with the support of machine learning. In this study, the stiction detection program was built by using the k-Nearest Neighbors (KNN) method and was implemented online so that the detection process could run quickly and in the real time. KNN is a supervised machine learning method that allows classifying data into certain groups based on the training data. Process variable (PV) and controller output (OP) data were processed by using time series feature extraction and feature scaling. The results were used to build the model. The obtained model was then implemented online by using sliding window and MQTT. From the results of the study, KNN model with k of 5, the window size data of 150, and the manhattan distance formula were obtained. This model had also been implemented in an online stiction detection program. The online test results also showed that the program could detect 32 of 34 SACAC control loops correctly.

Kata Kunci : stiction, katup kontrol, machine learning, k-nearest neighbors