Fault diagnosis of locomotive electro-pneumatic brake through uncertain bond graph modeling and robust online monitoring Gang Niu a,n , Yajun Zhao a , Michael Defoort b , Michael Pecht c a Institute of Rail Transit (IRT), Tongji University, Caoan 4800, Jiading, Shanghai 201804, China b LAMIH Automatic Control and Human Machine Systems Team, University of Valenciennes, Le Mont Houy, F59313 Valenciennes Cedex 9, France c CALCE Prognostics and Health Management Consortium, University of Maryland, College Park, MD 20742, USA article info Article history: Received 8 February 2014 Received in revised form 9 April 2014 Accepted 17 May 2014 Available online 6 June 2014 Keywords: Fault detection and diagnosis Uncertainty LFT-based bond graph modeling Auto-associative kernel regression Sequential probability ratio test abstract To improve reliability, safety and efficiency, advanced methods of fault detection and diagnosis become increasingly important for many technical fields, especially for safety related complex systems like aircraft, trains, automobiles, power plants and chemical plants. This paper presents a robust fault detection and diagnostic scheme for a multi- energy domain system that integrates a model-based strategy for system fault modeling and a data-driven approach for online anomaly monitoring. The developed scheme uses LFT (linear fractional transformations)-based bond graph for physical parameter uncer- tainty modeling and fault simulation, and employs AAKR (auto-associative kernel regression)-based empirical estimation followed by SPRT (sequential probability ratio test)-based threshold monitoring to improve the accuracy of fault detection. Moreover, pre- and post-denoising processes are applied to eliminate the cumulative influence of parameter uncertainty and measurement uncertainty. The scheme is demonstrated on the main unit of a locomotive electro-pneumatic brake in a simulated experiment. The results show robust fault detection and diagnostic performance. & 2014 Elsevier Ltd. All rights reserved. 1. Introduction To improve reliability, safety and efficiency, advanced methods of supervision, fault-detection and fault diagnostics become increasingly important for many technical processes—including safety related processes like aircraft, trains, automobiles, power plants and chemical plants [1]. A fault is defined as an unpermitted deviation of at least one characteristic of a variable from an acceptable behavior. Therefore, a fault may lead to a malfunction or failure of the system. Accordingly, a fault diagnosis system can be defined as a system that is used to “detect faults and diagnose their location and significance” [2]. A fault diagnosis system consists of the tasks of fault detection (FD), fault isolation (FI) and fault identification. FD makes a determination that either everything is operating within the specified normal range or that something has gone wrong. FI determines the kind and location of the fault, e.g., which component has degraded. Fault identification estimates the size, Contents lists available at ScienceDirect journal homepage: www.elsevier.com/locate/ymssp Mechanical Systems and Signal Processing http://dx.doi.org/10.1016/j.ymssp.2014.05.020 0888-3270/& 2014 Elsevier Ltd. All rights reserved. n Corresponding author. Tel.: þ86 21 6598 4712; fax: þ82 21 6598 4704. E-mail address: gniu@tongji.edu.cn (G. Niu). Mechanical Systems and Signal Processing 50-51 (2015) 676–691