An ILMI Approach to Robust Fault Detection Filter for a Drum Boiler System through a Time-Domain H-Index Norm Method Ladan Khoshnevisan * , Sadjaad Ozgoli ** , Majid Hassanzadeh Shojaei *** * Electrical engineering department, Faculty of Engineering, University of Tehran, Iran. l.khoshnevisan@ece.ut.ac.ir; khoshnevisan@ut.ac.ir ** Electrical and computer, engineering department, Tarbiat modares university, Tehran, Iran. ozgoli@modares.ac.ir *** Electrical and computer, engineering department, Tarbiat modares university, Tehran, Iran. majid_ee82@yahoo.com Abstract: Robust fault detection filter (RFDF) is mainly designed to detect faults in linear time-invariant (LTI) systems inherently exposed to external disturbances. H-index norm technique is one of the RFDF designing methods. The main idea of our study is to apply a continuous H-index method to a real boiler model which is made proper by adding an auxiliary direct channel. A high pass filter is augmented to raise the high frequency response. The H- index norm is maximized to distinguish between external disturbance and fault. The designed RFDF is continuous and can directly be implemented to the time-domain original continuous model. Furthermore, discretizing the continuous model may cause losing some information. Finally, the designed method is theoretically applied in a simulated model of a drum boiler operating in Synvendska Kraft AB Plant in Malmo, Sweden as a multivariable and strongly coupled system. It is illustrated that both sensor and actuator faults can robustly be detected as quickly as possible. To the best of authors' knowledge, this is the first time that the continuous H-index norm procedure is employed to detect actuator and sensor faults in a boiler model. Keywords: boiler, continuous H-index norm filter, fault detection, linear matrix inequality, robustness. 1. Introduction Boiler unit, which produces the required thermal energy for power generation, is one of the critical components of a power plant. An important characteristic of a properly boiler system is to control the drum water level. To gain such aim the drum water level should be sensed. Another important feature is to maintain a desired steam pressure at the outlet of the drum, which is satisfied by the combustion system. Combustion is a complex system with load disturbances and uncertainties. Because of the fault effects on the drum level and pressure, it may cause hazardous occurrences such as explosion in the power plant. As a result, faults should be detected as early as possible. A great deal of attention has been given to fault detection and isolation (FDI) model-based design in recent years due to higher performance, safety and stability standards. One of the most extensively studied schemes, among the various model-based methods, is observer-based approach. Robustness is an important aspect of model-based FDI systems [1]. The designed robust fault detection filter (RFDF) should involve two imperative capabilities. One of them is robustness against unknown inputs and uncertainties and the other one is being sensitive enough to the faults to be detected. Due to the coupling effects of the possible faults and disturbances, an appropriate trade-off between robustness and sensitivity should be considered for an FDI design. H- index norm technique is a suitable way to design an RFDF [2, 3]. Some other approaches for unknown input systems can be found in [4, 5 and 6]. Some authors proposed a mixed procedure such as H - /H  [7] in frequency domain. Some appropriate procedures such as H 2 -norm in frequency domain [8], H  -norm [9] and sliding mode [10] in time domain have been designed. However, the H  technique measures the maximum effect of an input on the output, contrary to main objective of fault detection. Some papers such as [11, 12] proposed an approach for discrete systems. In these methods the original model should be discretized and therefore, some information will be lost. The H-index norm is defined as the minimum singular value of a transfer function matrix in a particular frequency range [13, 14]. Maximizing this index can lead to distinguish between external disturbance and fault. The procedure presented here is adapted to be implemented in the continuous original models for practical purposes, while maintaining both robustness against disturbances and sensitivity to the faults simultaneously. Furthermore, this method proposes a linear matrix inequality (LMI) for H-index observer design through an iterative algorithm based on the technique presented in [15] which can easily be solved by LMI toolbox of MATLAB. In this procedure the finite frequency range is handled by frequency weighting. To the best of authors' knowledge, this is the first time that the continuous H-index norm technique has been used to detect actuator and sensor faults in a drum boiler model. In this paper an FDI approach is applied to a model of a real boiler located in Synvendska Kraft AB Plant in Malmo, Sweden. The rest of this paper is organized as follows: Section II presents the system description and preliminaries. In this step a system model including external disturbances is proposed. The H-index fault detection filter formulation is 922 ﻣﺘﻠﺐ ﺳﺎﯾﺖ MatlabSite.com MatlabSite.com ﻣﺘﻠﺐ ﺳﺎﯾﺖ