Fault detection and diagnosis of an industrial steam turbine using fusion of SVM (support vector machine) and ANFIS (adaptive neuro-fuzzy inference system) classifiers Karim Salahshoor a, * , Mojtaba Kordestani b , Majid S. Khoshro b a Department of Instrumentation and Automation, Petroleum University of Technology, Tehran, Iran b Department of Control Engineering, Islamic Azad University South Tehran branch, Iran article info Article history: Received 25 January 2010 Received in revised form 1 June 2010 Accepted 3 June 2010 Available online 30 August 2010 Keywords: ANFIS SVM OWA Fusion Fault diagnosis Steam turbine abstract The subject of FDD (fault detection and diagnosis) has gained widespread industrial interest in machine condition monitoring applications. This is mainly due to the potential advantage to be achieved from reduced maintenance costs, improved productivity and increased machine availability. This paper presents a new FDD scheme for condition machinery of an industrial steam turbine using a data fusion methodology. Fusion of a SVM (support vector machine) classifier with an ANFIS (adaptive neuro-fuzzy inference system) classifier, integrated into a common framework, is utilized to enhance the fault detection and diagnostic tasks. For this purpose, a multi-attribute data is fused into aggregated values of a single attribute by OWA (ordered weighted averaging) operators. The simulation studies indicate that the resulting fusion-based scheme outperforms the individual SVM and ANFIS systems to detect and diagnose incipient steam turbine faults. Ó 2010 Elsevier Ltd. All rights reserved. 1. Introduction Maintenance procedures are often provided by the equipment manufacturers to inhibit major fault occurrences, leading to possible shutdowns. The time intervals between these recommended procedures can be varied, based on a periodic maintenance schedule. One of the most expensive procedures [1] is dismantling, introducing the subsequent risk of creating problems such as vibration and leakage. Fault monitoring and diagnostic systems are efficient means of preventing regular and costly system maintenances. A fault repre- sents a deviation from the expected normal system behavior. Fault indicators can be elaborated on-line with available measurements. Fault detection comprises the conception of any relevant symptom from the fault indicators and the consequent evaluation of the time of fault occurrence. Fault diagnosis refers to fault-root discrimina- tion which can be based on an analytical model of the system, representing the normal system behavior in the absence of any fault [2]. This is by no means an easy task to be carried out, especially in non-linear dynamic systems [3], mainly due to the model imprecision, leading to difficulties in making a clear distinction between deviations made by model uncertainty and those imposed by a fault affecting the system or unknown distur- bances. This usually necessitates a trade-off to be considered between false alarm rate and missed detection rate. Following a proper fault diagnosis, recovery procedures can be implemented, resulting in fault tolerant control system [4]. For complex processes, however, obtaining a sufficiently precise analytical model is by no means an easy task to be done and hence other diagnostic approaches must be utilized [5]. Signal processing is a candidate approach to fault diagnosis when an analytical model is not a priori available [6]. Signal characteristics may be investigated either with time domain methods (e.g., correlation and mean-change), frequency domain methods (e.g., spectral analysis), or with more sophisticated methods like time-frequency or wavelet analysis [7]. However, the difficulty is how to correlate a change in some quantity (e.g., the signal mean, the spectrum, etc.) with the char- acteristic of a particular fault. Classification or pattern recognition approach presents a third way to deal with diagnostic objectives. This is mainly based on historical process data or expert knowledge about the system and its corresponding misbehaviors. Relevant symptoms are hence identified to be representative of each type of * Corresponding author. Tel.: þ98 91 23952203; fax: þ98 21 44214222. E-mail address: salahshoor@put.ac.ir (K. Salahshoor). Contents lists available at ScienceDirect Energy journal homepage: www.elsevier.com/locate/energy 0360-5442/$ e see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.energy.2010.06.001 Energy 35 (2010) 5472e5482