Robust State Estimation for Fouling Evolution in Batch Processes using the EM Algorithm

Ouyang Wu,Matthias Roth,Edmund Brekke,Lars Imsland,Stefan M Schneider,Ala E.F Bouaswaig

doi:10.1016/j.ifacol.2019.06.142

Ouyang Wu, Matthias Roth + Show 4 more

Open Access

https://doi.org/10.1016/j.ifacol.2019.06.142

Copy DOI

Abstract

Abstract Degradation such as fouling has significant negative impact on the efficiency of process manufacturing. Creating transparency about degradation is essential in the operations of the plants. This paper focuses on the state estimation of fouling in a multipurpose batch process, and the problem is formulated using an industrial case study. Due to the lack of a good model of fouling, the state estimation is integrated with parameter estimation using the expectation maximization (EM) algorithm, where the variational approximation method is employed for the intermediate robust state estimation in the E-step. The proposed state estimation is applied in the case study, which demonstrates the efficacy of the estimation of a recipe-independent key performance indicator for the batch-to-batch fouling.

Highlights

The multipurpose batch plant in the considered example produces many types of products according to the recipes
The polymerization starts with the inflow of the initiator, and the reactor temperature is controlled during the exothermic reaction using two parallel recirculation loops including pumps and heat exchangers
Fouling leads to an increased pressure drop over the heat exchanger resulting in a smaller recirculation flow

Summary

Introduction

The multipurpose batch plant in the considered example produces many types of products according to the recipes. The pressure drop measurement for overall batch is increasing, which matches the fouling trend from batch to batch; within an individual batch, the pressure drop, which is affected by the varying operating condition such as temperature, fails to indicate the actual fouling trend; a batch-to-batch fouling indicator is developed by sampling the pressure drop at the starting point of the reaction in each batch (denoted as asterisks in Fig. 2), where the operation conditions keep the same. We aim to develop state estimation for a recipe-independent fouling KPI, which can be further used in the modelling of fouling evolution. A set of linear function factors is used in the measurement equation to model the recipe effects on the fouling indicator

Objectives

Methods

Results

Conclusion