Abstract—Vibration-based structure health monitoring technique detects the damage by observing the change in dynamic characteristics of the structure. Change in dynamic characteristics due to operational and environmental variation may confuse with the change due to damage in the structure resulting false alarm of the damage. In SHM, data normalization technique can be used to suppress the adverse effect due to different operational and environmental variability. In this paper, a data normalization approach based on error prediction model is presented that estimates the residuals of the vibration feature due to damage. Damage is detected by processing the residual errors after applying Principal Component Analysis (PCA) on vibration features. The residual errors due to operational and environmental variabilities are optimally minimized through the best reconstruction of vibration features using an optimal number of principal components. The Variance of Reconstruction Error (VRE) is applied to obtain the optimum number of principal components for best reconstruction of vibration features. Relative standard deviation of the residual errors is used as damage index that quantifies the level of the damage in the structure. The proposed approach is validated on a benchmark problem of detecting damage in a three-story building under different operational and environmental variabilities. A comparative analysis is performed with previously reported work for damage detection to test the efficacy of the proposed algorithm.
 

Index Terms—auto-regressive model, principal component analysis, variance of reconstruction error, data normalization, damage index

Cite: Kundan Kumar, Prabir Kumar Biswas, and Nirjhar Dhang, "Statistical Damage Detection Approach in SHM Based on Error Prediction Model," International Journal of Structural and Civil Engineering Research, Vol. 5, No. 4, pp. 300-307, November 2016. doi: 10.18178/ijscer.5.4.300-307