Vibration based fault diagnosis for railway vehicle suspensions via a functional model based method: A feasibility study

Abstract

The design of a vibration based fault detection and isolation (FDI) unit that can tackle the combined problem of fault detection, isolation (or identification) and magnitude estimation (collectively known as fault diagnosis), in railway vehicle suspensions is presented. The unit is initially "trained" in a baseline phase based on data obtained from a simplified physics-based model of a railway vehicle suspension. Fault diagnosis is subsequently achieved in an inspection phase through a single, properly preselected, pair of vibration signals acquired from the vehicle, and a recently introduced data-based method, referred to as the functional model based method (FMBM), without resorting to the physics-based model of the baseline phase. The method's cornerstone is the novel class of stochastic ARX-type models capable of accurately representing a system in a faulty state for its continuum of fault magnitudes. Fault diagnosis feasibility in a railway vehicle suspension is demonstrated via Monte Carlo simulations using different types and magnitudes of faults in the physics-based model and generating vibration signals corresponding to the healthy and faulty suspension. Two vibration signals are used by the diagnosis unit: the track velocity profile and the vehicle body acceleration above the trailing airspring. Fault diagnosis based on the FMBM is effective in a compact and unified statistical framework accounting for experimental and modelling uncertainty through appropriate interval estimates and hypothesis testing procedures. The unit is shown to exhibit high sensitivity and accurate estimation of even very small fault magnitudes, to detect and isolate unknown faults for which it has not been trained, and to be robust to high measurement noise, car body mass variations, and varying track irregularity.

High-speed train suspension health monitoring using computational dynamics and acceleration measurements

“…In [15], a baseline Autoregressive with exogenous excitation (ARX) model is identified from experimental data. It represents the relation between the system input (the track excitation) and response (the accelerations) as an IIR filter with a white noise residual.…”

Rail vehicle suspension condition monitoring – approach and implementation

“…In [18] Multi-block Partial Least Squares approach is applied to build the statistical model using the historical normal operating data. A novel class of stochastic ARX-type models capable of representing a faulty state of vehicle is presented in [19].…”

IWSHM 2017: Vibration-based damage localization and estimation via the stochastic functional model–based method: An overview

“…A distinctly different approach, referred to as the functional model–based method (FMBM) , has been introduced by two of the present authors and co-workers. 30–34 Within it, the localization and magnitude estimation problems are, for the first time in the context of methods employing data - based models , treated as estimation (instead of classification) of the precise damage location coordinates and magnitude problems. A key novelty that paved the way for this development was the introduction of functional models (FMs) .…”

“…A key novelty that paved the way for this development was the introduction of functional models (FMs) . 30–36 These are special forms of autoregressive moving average with exogenous (ARMAX) excitation-type data-based models which generalize conventional counterparts by allowing them to depend on proper operating parameters, such as damage location coordinates and magnitude, referred to as scheduling variables (FMs may be viewed as generalizations of stochastic versions of linear parameter varying (LPV) models; the interested reader is referred to Sakellariou and Fassois 35,36 for details). These allow for the representation of the (generally partial) dynamics under any damage, at any location on the structural topology and of any magnitude, through a unique FM.…”

“…The method was originally postulated in a series of papers, 30,31,33 mainly for damage magnitude estimation using (scalar) FP-ARX models, and was validated on a replica of the Garteur SM-AG19 aircraft skeleton structure. In Sakellariou and Fassois, 32 it was applied to a simple, simulated, two-degrees-of-freedom (2-DOF) nonlinear system characterized by cubic stiffness using nonlinear (FP-NARX) models.…”