A hybrid electrical impedance tomography algorithm for detection of damages in composites
Electrical impedance tomography(EIT)is a non-radiative and low-cost imaging technique aiming to estimate the interior electrical properties of an object 1.Recently,EIT has been applied to damage detection for composites with their self-sensing ability enhanced by nanoscale fillers or surface mounted conductive skins 2.Nonetheless,EIT is typically an illposed inverse problem,regularization is usually required so as to obtain a stable solution.The widely used regularization is Tikhonov regularization which relaxes the ill-condition by balancing the original objective function and a smoothness penalty on the solution.The major difficulty of Tikhonov regularization lies in finding the optimal regularization parameter efficiently.To solve this problem,a Bayesian regularization(BR)has been developed with capability to adaptively determine the regularization parameter in a data-driven manner 3.In this work,Krylov subspace projection is combined with BR to form a hybrid algorithm for EIT to reconstruct the distribution of conductivity change caused by damages and thus quantitatively detect the damages in composites.In the hybrid algorithm,Krylov subspace projection is first employed to project the problem into a lower dimension subspace,then BR is used to determine the unknowns with an optimal regularization parameter value,leading to the reconstruction of the distribution of conductivity change with reduced computational cost.The proposed hybrid algorithm is experimentally verified.A glass fiber reinforced polymer(GFRP)plate with a surface attached carbon nanotube(CNT)film skin as shown in Figure 1(a)is considered.The dimensions of the CNT film are 10×10 cm2,and 20 electrodes connecting to a data acquisition system are equally spaced along its boundary.A constant DC current of 100 mA is injected into the film by adjacent electrode pairs sequentially and corresponding voltages are acquired from the rest of the electrode pairs.The test procedure is first performed under the pristine state to obtain a set of reference measurement data,then damages are introduced and the same testing procedure is performed under the damaged state.Figure 1(b)shows the GFRP plate with a circular hole damage whose diameter is 12 mm.Figures 2(a)shows the optimal regularization parameter adaptively determined through the data by BR,and Figure 2(b)shows the corresponding tomographic image of the reconstructed conductivity change.Figure 3(a)shows the singular values of the original sensitivity matrix and the projected sensitivity matrix in the hybrid algorithm(the dimensionality of the Krylov space is chosen as 80),and Figures 3(b)shows the corresponding tomographic image of the reconstructed conductivity change by the hybrid algorithm.It can be seen from Figure 2(a)and Figure 3(a),both algorithms can obtain the localized areas with significant conductivity reductions to well indicate the location and approximate size of the damage.However,the computational efficiency of the hybrid algorithm is much better than the BR algorithm: in a portable workstation with a dual-core Intel i7-4810MQ 2.5 GHz CPU and 16 GB memories,the hybrid algorithm only takes 0.13 seconds CPU time,but the Bayesian regularization algorithm takes 8.2 seconds CPU time,demonstrating the effectiveness of the proposed hybrid EIT algorithm for real-time or near real-time detection of damages.
G.Yan H.Sun Z.Yang
State Key Laboratory of Mechanics and Control of Mechanical Structures,College of Aerospace Engineer Department of Civil and Environmental Engineering,University of Pittsburgh,Pittsburgh,USA;Department
国际会议
The 7th World Conference on Structural Control and Monitoring(7WCSCM)(第七届结构控制与监测世界大会)
青岛
英文
1820-1822
2018-07-22（万方平台首次上网日期，不代表论文的发表时间）