464300 Power Plant Abnormal Condition Detection Using the Artificial Immune System Paradigm

Monday, November 14, 2016
Grand Ballroom B (Hilton San Francisco Union Square)
Ghassan Al-Sinbol1, Mario Perhinschi1 and Debangsu Bhattacharyya2, (1)Mechanical Engineering, West Virginia University, Morgantown, WV, (2)Department of Chemical Engineering, West Virginia University, Morgantown, WV

Power Plant Abnormal Condition Detection Using the Artificial Immune System Paradigm

Ghassan Al-Sinbol, Mario G. Perhinschi, Debangsu Bhattacharyya

            Modern power plants are expected to function safely and efficiently under both normal and abnormal operating conditions.  Due to the complexity, multi-dimensionality, and strong interaction among various sections of these plants, health monitoring requires comprehensive and integrated methodologies.  In this paper, the artificial immune system (AIS) paradigm is used to develop a detection scheme for abnormal operation of advanced power plants.

            The functionality of the biological immune system and its capability to discriminate between “self” (normal conditions) and “non-self” (abnormal conditions) inspired the AIS paradigm as a novel artificial intelligence technique with promising capabilities for abnormal condition detection.  An abnormal situation or failure affecting a dynamic system is considered similar to an antigen invasion.  Using a positive selection-type of algorithm, failures are declared or detected when the current configuration of “features” does not match with any configuration from an exhaustive set known to correspond to normal situations or the self.  Building the self requires a substantial amount of data collected at normal operational conditions; however, it does not require modeling of the system. 

            In this paper, a novel approach, denoted as ‘the partition of the universe’ approach, for self/non-self generation is presented.  In this approach, the n-dimensional feature space is divided into uniform partitions with predefined centers, shape, and size.  The raw self points are then tested against the universe partitions.  Self partitions are then identified and saved as integer strings that can be produced and used with reduced computational effort. 

            An abnormal condition detection scheme is designed based on a positive-selection-type of algorithm in conjunction with a dendritic cell (DC) mechanism that is expected to allow extension towards abnormal condition identification and evaluation.  The artificial DC is a computational module inspired from the interaction between the innate and adaptive immune systems.  The proposed artificial DC mechanism is expected to provide the detection outcome based on the current and past discrimination results and overcome any imperfections in the definition and generation of the self.  The general flowchart of the detection process is presented in Figure 1 and the flowchart of the proposed abnormal condition detection scheme using the partition of the universe approach is presented in Figure 2.

            The proposed approach is demonstrated with promising results using a rigorous model of an acid gas removal (AGR) unit as part of the integrated gasification combined cycle power plant developed in Dynsim® environment.  A total of 150 features was selected to build the self/non-self of the AGR unit, including pressure, temperature, flow rate, and composition measurements across the unit.  The AGR unit is divided into 22 subsystems and lower dimensional projections of the self are considered for each subsystem within the hierarchical multi-self strategy.  For the purpose of demonstrating the operation of the proposed detection scheme, a limited number of abnormal conditions that include deposition of solids, such as flyash, and leakages in the pipes or equipment items has been considered.  These abnormal conditions affect five main sub-systems characterized by sixty features. 

            The proposed detection scheme provides excellent performance with high detection rates and zero false alarms for all cases considered.  The detection rates vary between 94.6% and 99.3% with detection times between 2.5 and 10.5 seconds.

Figure 1.  AIS-based Abnormal Condition Detection

Figure 2.  Abnormal Condition Detection Using the Partition of the Universe Approach


Extended Abstract: File Not Uploaded
See more of this Session: Interactive Session: Systems and Process Control
See more of this Group/Topical: Computing and Systems Technology Division