Abstract

An experiment was implemented using the electro-mechanical impedance (EMI) health monitoring technology for a framed structure and the conductance data were accordingly obtained for various damage severities. The compressed data by compression sensing theory were used to character the structural damages instead of the raw ones, and were further studied based on the principal component analysis (PCA). The obtained principal components were then employed to be the input parameters of the BP artificial neural network (ANN). Results showed that the transmission bandwidth and storage space of the EMI data were only 40% of the original ones using the present method. The neural network could identify the appearance of damages and could further classify the damage severities quantitatively using the principal components of the compressed conductance. Introduction Electro-mechanical impedance method has been demonstrated to be a potentially powerful structural damage detection technique. The identification of cracks, damages in metallic structures, loosening connections of pipelines, and delaminations in composites structures has been investigated by many researchers [1,2]. For an online real-time EMI structural health monitoring system, the measured data acquisition and processing are very important. The EMI signals should be transmitted timely to the system processor, which can then make a proper judgment for damage diagnose. However, in the practical engineering applications, the obtained EMI data are very large. In order to achieve efficient transmission and timely diagnosis function, the raw data should be compressed to reduce the computer memory. On the other hand, the compression sensing theory [3,4] has attracted much attention in the field of signal processing. In this paper, the compression sensing technique will first be used for pre-treatment of measured EMI signatures to reduce the dimension of data. The basic idea of this method is the global observation to the sparse signal far below the Nyquist sampling rate rather than local sampling [5]. A principal component analysis (PCA)-based data reduction technique is further applied to the compressed EMI data. The compressed EMI data, represented by their projection onto the most significant principal components, are then used as ANN input variables instead of the pristine EMI data. Experiment and discussion A steel frame with bolt joints, as shown in Figures 1, is investigated. The piezoelectric patches are bonded on the critical section of certain joint of the frame and are connected with impedance analyzer 4294A by two wires. Damages are simulated by completely loosening bolts. In this experimental case, five conditions are taken into account: (1) Pristine; (2) loosening the No.1 bolt; (3) loosening the No.1 and No.2 bolts; (4) loosening the No.1, No.2 and No.3 bolts; (5) loosening the No.1, No.2, No.3 and No.4 bolts. Figure 2 displays the real (G) and imaginary part (B) of admittance for the experimental case. The piezoelectric admittance changes obviously with the increasing number of loosening bolts. This is because the structure damping and the dissipation of energy increase and conversely the structural International Conference on Applied Science and Engineering Innovation (ASEI 2015) © 2015. The authors Published by Atlantis Press 313 stiffness of steel frame decreases. Therefore, EMI technology can identify structural damage well. However, Figure 2 just reflects the existence of damage qualitatively and can not assess damage severities quantitatively. Besides, in the practical applications, signal sampling points are numerous, the direct acquisition of signal transmission and storage to test system is very difficult. Therefore, this paper will put forward not only compress memory but also quantitative damage assessment method, which is based on compression sensing of EMI signatures and ANN structural damage identification method. Figure 1. Experimental set up Figure 2. Admittance signatures for different damage severity The compression sensing of conductance signal Set x is one dimensional signal, its length is N and the sparse degree is K. Φ is the two-dimensional matrix M×N (M < N). x y Φ = is one dimensional measurements value for the length of M. Compression sensing problem is to solve the underdetermined equations y x = Φ , and get the original signal x based on known measured values y and the observation matrixΦ . It is necessary to solve the optimization problem as follows: y x t s x x = Φ = . min arg 0 (1) It has showed that the number of observations is four times or above than that of the sparse degree, it can be realized that the signals are reconstructed without any distortion [3,4]. In the present experiment, five operating conditions are considered and meanwhile 3 sets data including conductance, susceptance and the modulus of admittance are taken to be the input parameters. Thus, 15 sets data should be taken into account. Therefore, once the conductance signal sparse degree K is known, only 4×K time Gaussian random observations are necessary:

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call