Razvoj i primjena softverskog senzora na postrojenju za proizvodnju propilena

master's thesis

• Title (english): Development and application of a soft sensor at a propylene production plant
• Author: Ines Martić
• Mentor: Željka Ujević Andrijić (mentor)
• Committee member: Željka Ujević Andrijić (predsjednik povjerenstva)
• Committee member: Nenad Bolf (član povjerenstva)
• Committee member: Miroslav Jerković (član povjerenstva)
• Granter: University of Zagreb; Faculty of Chemical Engineering and Technology; Zagreb
• Defense date and country: 2023-11-29, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Chemical Engineering

Abstract

U ovome radu prikazan je razvoj modela softverskih senzora za kontinuirano praćenje sadržaja propilena u rafinerijskom postrojenju propan/propilen splitter. Dobiveni podaci ulaznih varijabli usklađeni su vremenski sa izlaznom varijablom te su obrađeni različitim metodama predobrade. U programskom jeziku Python izrađeni su modeli softverskih senzora primjenom neuronskih mreža s višeslojnim perceptonom (engl. Multi-layer percepton, MLP) i neuronskih mreža s dugotrajnim kratkoročnim pamćenjem (engl. long short-term memory, LSTM). Pri razvoju MLP modela analizirani su hiperparametri modela kao što su: broj neurona u skrivenom sloju i utjecaj vrste prijenosne funkcije (sigmoid, tanh, ReLU i ELU) na kvalitetu razvijenih neuronskih mreža. Pri razvoju LSTM modela se dodatno ispitivao i broj vremenskih koraka u prošlost te broj LSTM jedinica. Oba modela dala su visoke i slične vrijednosti koeficijenata korelacije te niske pogreške između modeliranog sadržaja propilena i realnih podataka sa postrojenja te se mogu primijeniti u rafinerijskom informacijskom sustavu. Analiza trenda i histograma pogreške razvijenih modela dodatno ukazuje na zadovoljavajuću kvalitetu razvijenih modela. Primjenom softverskih senzora smanjuje se potreba za korištenjem skupe mjerne analitičke opreme i opreme koja se često kvari, te je u konačnici moguće potpuno zamijeniti postojeće analizatore u procesu. Primjenom softverskih senzora na postrojenju unaprijedit će se automatsko vođenje procesa što će utjecati na stabilnije odvijanje procesa i veću kvalitetu konačnog proizvoda.

Abstract (english)

This paper presents the development of soft sensor models for continuous monitoring of propylene content in rafinery propane/propylene splitter. The goal of this paper is to develop soft sensor models based on given experimental data about refinery process of propane and propylene separation, validate and compare developed models. The development of models began with the preprocessing of the collected measurement data. Soft sensor models were created in the Python programming language using the methods of multilayer percepton networks and long short-term memory networks. During development of multilayer percepton networks model it was tested how hyperparameters such as number of neurons in hidden layers and activation function (sigmoid, tanh, ReLU i ELU) affect on quality of the developed neural networks. During the development of LSTM models, additional testing of the number of time steps into the past and the number of LSTM units on model quality were examined. A statistical analysis of the results was carried out, which shows that both types of models give high correlation values of model data with real propylene content data, so both types of models can be applied in the refinery information system. Error trend analysis and error histogram of the developed models additionally indicate the satisfactory quality of the developed models. By applying the developed soft sensors it is possible to decrease the need for high cost measurement equipment and equipment that breaks down frequently, and ultimately completely replace the existing analyzers in the process and improve the automatic management of the process. Application of software sensors will improve automatic process control, which will result in more stable process management at the plant and higher quality of the final product.

• Keywords: softverski senzor; neuronske mreže; višeslojni percepton; mreže s dugotrajnim kratkoročnim pamćenjem; propilen
• Keywords (english): software sensor; neural networks; multi-layer percepton; long short-term memory networks; propylene
• Language: croatian
• URN:NBN: urn:nbn:hr:149:297122
• Study programme: Title: Chemical Engineering - Graduate study; Study programme type: university; Study level: graduate; Academic / professional title: sveučilišni/a magistar/magistra inženjer/inženjerka kemijskog inženjerstva (sveučilišni/a magistar/magistra inženjer/inženjerka kemijskog inženjerstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2023-12-11 14:39:08