MACHINE LEARNING BASED SOFT SENSOR DESIGN FOR OPTIMIZATION IN REALTIME PETROCHEMICAL PROCESSES

Conference paper > Summary Text · 2024

Authors: Elshan Mehdiyev, Sevim YOLCULAR KARAOĞLU, Burak Alakent

3. BİLSEL INTERNATIONAL TRUVA SCIENTIFIC RESEARCHES AND INNOVATION CONGRESS, Türkiye, Çanakkale, International, Verbal, Summary Text, 2024-2024, Scopus, vol 1, pp 726-pdf807-726-pdf807

Summary

<p class="ql-align-justify"> Effective&nbsp;process&nbsp;monitoring&nbsp;is&nbsp;crucial&nbsp;for&nbsp;achieving&nbsp;sustainability.&nbsp;Traditional&nbsp;hardware&nbsp;sensors,&nbsp;while&nbsp;common,&nbsp;are&nbsp;often&nbsp;inadequate&nbsp;due&nbsp;to&nbsp;harsh&nbsp;working&nbsp;environments,&nbsp;measurement&nbsp;delays,&nbsp;and&nbsp;high&nbsp;costs.&nbsp;These&nbsp;limitations&nbsp;are&nbsp;particularly&nbsp;problematic&nbsp;for&nbsp;measuring&nbsp;product&nbsp;quality&nbsp;variables,&nbsp;which&nbsp;are&nbsp;typically&nbsp;assessed&nbsp;through&nbsp;offline&nbsp;laboratory&nbsp;analyses.&nbsp;To&nbsp;overcome&nbsp;these&nbsp;challenges,&nbsp;researchers&nbsp;have&nbsp;developed&nbsp;soft&nbsp;sensors-mathematical&nbsp;models&nbsp;that&nbsp;estimate&nbsp;hard-to-measure&nbsp;quality&nbsp;parameters&nbsp;in&nbsp;real-time.&nbsp;My&nbsp;scientific&nbsp;research&nbsp;focuses&nbsp;on&nbsp;predicting&nbsp;the&nbsp;ring&nbsp;loss&nbsp;value&nbsp;in&nbsp;Isomerization&nbsp;reactions&nbsp;using&nbsp;online&nbsp;measurable&nbsp;process&nbsp;parameters,&nbsp;thereby&nbsp;optimizing&nbsp;production&nbsp;efficiency&nbsp;in&nbsp;the&nbsp;Aromatics&nbsp;plant.&nbsp;By&nbsp;maintaining&nbsp;the&nbsp;ring&nbsp;loss&nbsp;within&nbsp;optimal&nbsp;ranges,&nbsp;valuable&nbsp;product&nbsp;yield&nbsp;can&nbsp;be&nbsp;maximized&nbsp;while&nbsp;minimizing&nbsp;losses.&nbsp;The&nbsp;methodology&nbsp;includes&nbsp;designing&nbsp;a&nbsp;soft&nbsp;sensor,&nbsp;preparing&nbsp;data,&nbsp;and&nbsp;optimizing&nbsp;the&nbsp;unit&nbsp;using&nbsp;Aspen&nbsp;HYSYS,&nbsp;ultimately&nbsp;applying&nbsp;the&nbsp;soft&nbsp;sensor&nbsp;to&nbsp;the&nbsp;Aromatics&nbsp;plant.&nbsp;Regressions&nbsp;developed&nbsp;using&nbsp;historical&nbsp;data&nbsp;and&nbsp;machine&nbsp;learning&nbsp;techniques&nbsp;in&nbsp;Python&nbsp;will&nbsp;predict&nbsp;process&nbsp;variables,&nbsp;ensuring&nbsp;the&nbsp;ring&nbsp;loss&nbsp;value&nbsp;remains&nbsp;within&nbsp;optimal&nbsp;limits.&nbsp;Integrating&nbsp;this&nbsp;AI-powered&nbsp;soft&nbsp;sensor&nbsp;into&nbsp;real-time&nbsp;petrochemical&nbsp;processes&nbsp;will&nbsp;facilitate&nbsp;better&nbsp;process&nbsp;optimization.&nbsp;In&nbsp;conclusion,&nbsp;AI-based&nbsp;soft&nbsp;sensors&nbsp;represent&nbsp;a&nbsp;pivotal&nbsp;advancement&nbsp;in&nbsp;modern&nbsp;process&nbsp;industries.&nbsp;They&nbsp;enable&nbsp;precise&nbsp;real-time&nbsp;estimation&nbsp;of&nbsp;quality&nbsp;parameters,&nbsp;enhancing&nbsp;process&nbsp;monitoring,&nbsp;control,&nbsp;and&nbsp;decision-making.&nbsp;This&nbsp;leads&nbsp;to&nbsp;reduced&nbsp;energy&nbsp;consumption,&nbsp;lower&nbsp;emissions,&nbsp;and&nbsp;a&nbsp;minimized&nbsp;environmental&nbsp;footprint,&nbsp;aligning&nbsp;industrial&nbsp;practices&nbsp;with&nbsp;sustainable&nbsp;development&nbsp;goals.</p><p><strong style="color: black;">Keywords:&nbsp;&nbsp;</strong><span style="color: black;">Soft&nbsp;sensors,&nbsp;AI-based&nbsp;soft&nbsp;sensors,&nbsp;Aspen&nbsp;HYSYS,&nbsp;Python,&nbsp;Machine&nbsp;Learning.</span></p>

View details