Estimation of fouling thickness in low density polyethylene tubular reactor process using soft sensor model

• Main Authors: Fakhrony Sholahudin Rohman, Fakhrony Sholahudin Rohman, Muhammad, Dinie, Azmi, Ashraf, Murat, Muhamad Nazri
• Format: Conference or Workshop Item
• Published: 2023
• Subjects: TP Chemical technology
• Online Access: http://eprints.utm.my/108221/; http://dx.doi.org/10.1063/5.0149823
• Institution: Universiti Teknologi Malaysia

In this research, a comparison between the Linear support vector machine (LSVM) and Quadratic support vector machine (QSVM) technique for estimating fouling layer thickness in Low density polyethylene (LDPE) tubular reactor process is performed. In practice, fouling is a form of polymer layer deposited inside the reactor wall, which is caused during thermodynamically driven phase separation of polymer (i.e., LDPE) and monomer (i.e., ethylene). The formation of fouling can reduce the heat transfer through the reactor's wall due to the low thermal conductivity of the deposited material. Thus, fouling can reduce the LDPE tubular reactor productivity by affecting the final polymer quality. However, the study of fouling formation for the LDPE tubular reactor process is hardly reported in the literature since it is hard to measure. Therefore, a machine learning technique, namely SVM, has been selected as soft sensor model to estimate the fouling thickness build-up inside LDPE tubular reactor, specifically in the cooling zone. In order to develop the SVM model, a set of fouling data is generated from a combination of LDPE tubular reactor model simulation using Aspen Dynamic and fouling build-up equation. In order to improve the soft sensor model, input selection, the Pearson correlation coefficient (PCC) method is implemented. Based on PCC analysis, polymer density, heat transfer coefficient, and reactor temperature in the respective cooling zone were considered as inputs for the developed soft sensor model. Based on the modeling results, the QSVM model performed better (with an R2 of 0.99) than the LSVM model (with an R2 of 0.97). Thus, the application of soft sensor model is using QSVM for the LDPE tubular reactor process is justified.