PERANCANGAN PENGONTROL ADAPTIF BERBASIS ALEURO FUZZY PADA DISTILASI KOLOM JENIS BUBBLE CAP KONTINYU

• Main Author: Shoelarta, Shoerya
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/1972
• Institution: Institut Teknologi Bandung

Bubble cap column distillation is one of the mostwidely used separation <br /> process units. The separation is based on boiling point difference between <br /> components. Ethanol-aqua mixture is a difficult mixture to separate for <br /> achiving high purity (>95%) continously due to azeotropic property. The <br /> existing controllers for the unit in Pilot plant laboratory at Politeknik <br /> Negeri Bandung are Siemens SC 135 PID controllers which are relativelly <br /> modern, fully featured but do not quite fulfill the expectation. <br /> In this research, an Adaptive Neuro Fuzzy Inference Sytem (ANFIS) is <br /> applied by means of a hybrid learning algorithm using Error Back <br /> Propagation (EBP) and Least Square Estimator (LSE). The ANFIS <br /> learning results in a Fuzzy Inference System (FIS) matrix with 64 rules and <br /> reasonably small error of 0.004 after 40 epochs. <br /> This FIS matrix is then used as plants model, which is a function transfer as <br /> well. An inverse dynamic controller is applied consisting of a matrix of 1 <br /> input that is error and 6 outputs. The simulated closed loop will give rise <br /> relatively fine respons, which appropriate tuning of membership functions. <br /> The residual error is less than 1%, although a 3 minute settling time is <br /> inevitable. <br /> A modified controller with 2 inputs namely error and rate of error gives rise to <br /> even better responds, under 0.5 % of residual error and 3 minute settling time <br /> which are significantly better than that of existing PID controllers in real <br /> practice that are 1-2 % and 10-11 minute respectively. Altough the PID <br /> respons at 1 minute, faster than the simulated plant at 2 minutes. <br /> Key words : distillation, separation, neuro_fuzzy, inverse dynamics, error