Design and simulation of heat exchangers and three-phase separators for efficient desulfurization in petrochemical processes

This paper presents a thorough approach to enhancing desulfurization processes in petrochemical plants through machine learning-enhanced design of heat exchangers and three-phase separators. We combined ASPEN ASPEN v14 process modeling with advanced regression analysis to identify optimal equipment configurations and operational parameters. The desulfurization process involves heating the feed to 205 °C, followed by phase separation in a three-phase separator and subsequent sequential flashing. We developed multiple linear regression models to improve heat exchanger performance, achieving a R 2 value of 0.9114. The regression analysis revealed that tube length and baffle spacing are the essential design parameters. The altered heat exchanger achieved an overall heat transfer coefficient of 543.58 W/m 2 °C, indicating a 6.1 % improvement relative to the baseline design. Through systematic parameter variation and regression modeling, we determined the optimal operating conditions: a valve outlet pressure of 80 kPa, an air cooler outlet temperature of 60 °C, and a fired heater outlet temperature of 205 °C. The three-phase separator, with a diameter of 4 m and a length of 10 m, was designed in compliance with API 12J standards, with redesigned internals for improved separation efficiency. The system demonstrated outstanding efficacy, achieving 99.6 % sulfur removal and reducing H 2 S concentration from 35,534 ppm to only 11.59 ppm. The optimization method driven by machine learning enabled systematic equipment sizing and revealed substantial design trade-offs. The conclusive design of our heat exchanger, consisting of 22 tubes configured in a 1–2 pass arrangement with a baffle spacing of 0.0548 m, meets all thermal and hydraulic standards as per TEMA regulations. This paper introduces a validated methodology for desulfurization system design that integrates process simulation with machine learning optimization, yielding quantifiable performance improvements over conventional design approaches.