Identifikasi dan Implementasi Self-Tuning System Berbasis Radial Basis Function pada Kendali PID untuk Reaction Wheel

Abstract

Reaction wheel merupakan aktuator pada sistem kendali sikap satelit yang menghasilkan torsi berdasarkan prinsip kekekalan momentum sudut. Pengendali PID konvensional memerlukan tuning manual dan kurang adaptif terhadap perubahan kondisi operasi maupun gangguan. Penelitian ini bertujuan mengidentifikasi karakteristik dinamik reaction wheel, merancang pengendali self-tuning PID berbasis Radial Basis Function Neural Network, serta mengevaluasi kinerjanya. Identifikasi sistem dilakukan menggunakan model Hammerstein–Wiener dan Nonlinear AutoRegressive with eXogenous Input untuk hubungan PWM–arus dan arus–kecepatan. Hasil identifikasi menghasilkan nilai best fit sebesar 97,11% dan 99,14%. Sistem kendali menggunakan struktur cascade dengan loop dalam berupa pengendali PI arus dan loop luar berupa pengendali RBF-PID kecepatan. Pada pengujian nominal, pengendali menghasilkan settling time sebesar 93,5 detik, overshoot sebesar 44,55 RPM, dan arus puncak sebesar 1026,85 mA. Dibandingkan pengendali PI konvensional, metode yang diusulkan memperpendek settling time sebesar 6,5%, menurunkan overshoot sebesar 25,7%, serta mengurangi arus puncak sebesar 13%. Pada gangguan beban torsi sebesar 0,2 Nm, sistem menghasilkan RMSE sebesar 43,24 RPM dan 44,40 RPM pada kondisi noise pengukuran. Implementasi pada mikrokontroler STM32L4R5ZI menghasilkan waktu eksekusi rata-rata 75 µs dengan beban CPU 8,3%.

A reaction wheel is an actuator in satellite attitude control systems that generates torque based on the conservation of angular momentum. Conventional PID controllers require manual tuning and lack adaptability to changing operating conditions and disturbances. This study identifies the dynamic characteristics of a reaction wheel, designs a self-tuning PID controller based on a Radial Basis Function Neural Network, and evaluates its performance. System identification was performed using Hammerstein–Wiener and Nonlinear AutoRegressive with eXogenous Input models for the PWM–current and current–speed relationships. The identified models achieved best-fit values of 97.11% and 99.14%, respectively. The proposed control system employed a cascade structure with an inner PI current loop and an outer RBF-PID speed loop. Under nominal conditions, the controller achieved a settling time of 93.5 s, an overshoot of 44.55 RPM, and a peak current of 1026.85 mA. Compared with a conventional PI controller, the proposed method reduced settling time by 6.5%, overshoot by 25.7%, and peak current by 13%. Under a 0.2 Nm torque disturbance, the system achieved RMSE values of 43.24 RPM and 44.40 RPM under measurement noise. Implementation on an STM32L4R5ZI microcontroller achieved an average execution time of 75 µs with an 8.3% CPU load.