Vibration Based Fault Detection in a Time-Varying Link Structure via Non-Stationary FS-VTAR Models

The problem of vibration based fault detection in a time-varying, extendable, prismatic link structure is addressed. The structure is extended from its fully retracted to its fully expanded position at a constant speed, while it is subject to broadband random force excitation. The faults considered are of various types and occurrence locations, while their detection is based solely on the non-stationary vibration response signals acquired during operation. A novel Functional Series Vector Time Dependent AutoRegressive (FS-VTAR) method combined with an appropriate statistical decision framework is postulated for tackling the problem, and its performance is experimentally assessed. 316 IOMAC'09 – 3 International Operational Modal Analysis Conference Poulimenos and Fassois (2004) who use an input-output model and a statistical decision making framework for the on-line fault identification in a “bridle-like” laboratory structure. Also Zhan and Jardine (2005) utilize a time-varying autoregressive model in order to detect the presence of a fault in rotating machinery. Wang et al. (2008) extended this work by using classification for the purpose of fault identification based on a feature vector extracted from parametrically estimated spectra of the non-stationary vibration response signal. The aims of the present study are: (a) The introduction of a statistical (parametric) model based method for vibration based fault detection in inherently non-stationary structures, and (b) the assessment of its effectiveness on a laboratory structure. The proposed method is based on novel non-stationary Functional Series Vector AutoRegressive (FS-VTAR) modelling and a proper statistical decision making scheme. 2 THE EXPERIMENTAL SETUP, THE FAULTS, AND THE EXPERIMENTS 2.1 The experimental setup The extendable prismatic link structure is shown in Fig. 1. It consists of a fixed-free outer frame and an inner beam sliding in it as it is driven by an AC motor (Sanyo Denki AC servomotor P30B06040DXS). The structure’s main dimensions are summarized in Table 1. Table 1: Main dimensions for the extendable prismatic link structure. Length (mm) Width (mm) Height (mm) Frame 1200 125 170 Beam 1070 45 90 The link structure is subject to zero-mean and Gaussian random force excitation which is vertically exerted via an electromechanical shaker (MB Dynamics Modal 50A, maximum loading 225 N) equipped with a stinger. The structural response is measured at six selected locations (Fig. 1; locations 16) via PCB 740B02 dynamic strain gauges (frequency range 0.5 – 100,000 Hz), while the measured strain signals are conditioned and subsequently driven into a SigLab 20-42 data acquisition module (two 16-bit D/A channels, four 20-bit simultaneously sampled A/D channels, analog 4-order quasi elliptic anti-aliasing filters). 2.2 The faults Two different types of faults are considered (Fig. 2). In the first case (Fault A), one of the four bolts which support the structure at its fixed end is loosened from 25 Nm to 5 Nm, while in the second (Fault B) case one of the four rolling guides at the free end of the structure is removed. Figure 1: The prismatic link structure and the experimental setup: (a) Photo, and (b) schematic diagram.