Pregled bibliografske jedinice broj: 1263419
Data-driven estimation of critical quality attributes on industrial processes
Data-driven estimation of critical quality attributes on industrial processes // 19th Ružička Days “Today Science – Tomorrow Industry”
Vukovar, Hrvatska, 2022. str. 98-111 (poster, recenziran, cjeloviti rad (in extenso), znanstveni)
CROSBI ID: 1263419 Za ispravke kontaktirajte CROSBI podršku putem web obrasca
Naslov
Data-driven estimation of critical quality attributes on industrial processes
Autori
Ujević Andrijić, Željka ; Herceg, Srečko ; Bolf, Nenad
Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni
Izvornik
19th Ružička Days “Today Science – Tomorrow Industry”
/ - , 2022, 98-111
Skup
International scientific conference 19th Ružička Days “Today Science – Tomorrow Industry”
Mjesto i datum
Vukovar, Hrvatska, 21.09.2022. - 23.09.2022
Vrsta sudjelovanja
Poster
Vrsta recenzije
Recenziran
Ključne riječi
data-driven model ; process monitoring ; MLP artificial neural network ; isomerisation process
Sažetak
Large data sets generated and stored by continuous monitoring of process variables in process plants enable the development of data-driven mathematical models of a process. While such models do not provide detailed insight into the process, they do provide a feasible description of the nonlinear complex process dynamics. Data-driven models are often used to estimate important process characteristics in real-time. This paper presents an industrial case study of data-driven models based on multilayer perceptron (MLP) artificial neural networks (ANN) for the isomerization process of light naphtha in refineries. Models were developed for estimating critical quality characteristics of the product. Special attention was paid to the selection and analysis of data sets for different time periods to capture different process conditions. Optimal models were achieved by changing the types of learning function and transfer function, as well as the number of neurons in the hidden layer. Models based on MLP neural networks have shown better generalization capabilities compared to previously developed polynomial linear and nonlinear models for isomerization processes. This makes the developed MLP network models suitable for application to such a process, especially in the area of advanced process control.
Izvorni jezik
Engleski
Znanstvena područja
Kemijsko inženjerstvo
POVEZANOST RADA
Ustanove:
Fakultet kemijskog inženjerstva i tehnologije, Zagreb
