Date

2017

Document Type

Thesis

Degree

Master of Science

Department

Mechanical Engineering

First Adviser

Alparslan Oztekin

Abstract

Ammonium bisulfate (ABS) is an acidic deposit that can form on the metal elements of air preheaters in power boilers, leading to unit operational issues. As a byproduct of the Selective Catalytic Reduction (SCR) systems for nitrogen oxide (NOx) emissions control, ABS could result in unit efficiency deterioration, even unit outage. ABS formation temperature is an important factor in controlling the issues associated with ABS fouling problems. If the ABS formation temperature could be monitored, the ABS deposition location could be identified. Subsequently, preventative actions could be taken to avoid ABS fouling to develop into a serious operational problem, such as air preheater plugging. This study deals with indirect predictive models of ABS formation temperature. Five models were developed based on data mining technologies, using actual power plant data. Data composed of 14,230 samples, from 49 variables were used in the study. In the modeling, Principal Component Analysis (PCA) and Sensitivity Analysis (SA) were used to reduce the number of variables in the data set. K-Means Clustering (KMC) was also employed to compress training samples. Neural Networks (NN) and Support Vector Machine (SVM) were used for data modeling. Model results were validated with ABS formation temperatures measured with an ABS dew-point probe. A SA was performed to determine the impact of individual variables on the ABS formation process. It was found that four unit variables: SO2 stack concentration, SCR gas outlet temperature, SCR inlet NOx concentration and dilution skid ammonia flow, can provide a good representation of the data set for ABS formation temperature prediction. The most accurate predictive model consists of a sequence of KMC and SVM. This approach can predict ABS formation temperature within a 9% error from the physical measurement.

Share

COinS