Early Detection of Boiler Leakage in a Combined Cycle Power Plant Using an Auto Associative Kernel Regression Model

This paper presents the results of applying a data-driven condition-based monitoring system for the fault detection of a boiler leakage in a Combined Cycle Power Plant (CCPP). An auto associative kernel regression model is developed using normal process data and tested with faulted data to determine the earliest warning of the boiler leakage. Automatic variable grouping, which uses the linear correlations among the available thirty sensors, is employed to obtain optimal groupings to be used in model development. Several models were developed, optimized and compared. A logic test was used for fault detection and this test produced alarms in the region were the leak was later confirmed to have occurred. Comparison of these results with those of a physics-based analysis also confirmed the accuracy of the models in the early detection of the leakage.Copyright © 2013 by ASME and Alstom Technology