Sistem kolom penstabil pada unit distilasi minyak mentah (crude distillation unit) rentan terhadap fluktuasi termal yang berisiko menurunkan kemurnian produk. Ketiadaan model dinamis dan risiko pengujian langsung pada unit beroperasi menuntut metode penalaan yang andal dan tidak mengganggu proses.
Penelitian ini meningkatkan performa pengendali suhu XXTIC01 melalui pendekatan data-driven berbasis data historis Distributed Control System (DCS) menggunakan identifikasi sistem model Box-Jenkins yang memisahkan dinamika aktualdari gangguan, validasi model melalui tingkat kecocokan dan analisis residual, reduksi model menjadi bentuk FOPDT (First Order Plus Dead Time) dengan algoritma optimasi, serta penalaan berbasis Skogestad Internal Model Control (SIMC) dengan algoritma autodetuning dan filter titik acuan.
Hasil pengujian menunjukkan model Box-Jenkins memiliki tingkat kecocokan 86,10% dan lolos seluruh uji residual. Hasil reduksi model menghasilkan parameter FOPDT K = ?2,0733, ? = 147,18 menit, dan ? = 1,0 menit dengan tingkat kecocokan 81,73% terhadap model Box-Jenkins. Hasil penalaan dengan SIMC mendapatkan parameter Kp = ?11,8314 dan Ti = 24,0 menit yang mampu menekan overshoot dari 8,76% menjadi 1,89%, mempercepat settling time dari 119 menit
menjadi 53 menit, dan meningkatkan kinerja sistem kontrol akhir sebesar 29,68%.
The stabilizer column system in the Crude Distillation Unit (CDU) is susceptible to thermal fluctuations that may reduce product purity. The absence of a dynamic model and the risk of conducting a direct step test on an operating unit require a reliable tuning method that does not disturb the process.
This study improves the performance of the XXTIC01 temperature controller through a data-driven approach based on historical Distributed Control System (DCS) data using Box-Jenkins system identification, which separates the actual process dynamics from disturbances, model validation through fit percentage and residual analysis, model reduction into a FOPDT (First Order Plus Dead Time) form using an optimization algorithm, and tuning based on Skogestad Internal Model Control (SIMC) with an autodetuning algorithm and setpoint filter.
The results show that the Box-Jenkins model achieved a fit percentage of 86.10% and passed all residual tests. The model reduction produced FOPDT parameters of K = ?2.0733, ? = 147.18 minutes, and ? = 1.0 minute, with a fit percentage of 81.73% against the Box-Jenkins model. SIMC tuning obtained controller parameters of Kp = ?11.8314 and Ti = 24.0 minutes, which reduced the overshoot from 8.76% to 1.89%, accelerated the settling time from 119 minutes to 53 minutes, and improved the final control system performance by 29.68%.
Kata Kunci :
kolom penstabil, identifikasi sistem, Box-Jenkins, reduksi model, SIMC
