A Method Of Structural Damage Diagnosis Based On JS Divergence

Vibration-based damage detection technique is a kind of method which has the potential to develop into a long-term real time monitoring technique in the circumstances that it imposes minimum effect on the operation to the structures. The structural response under ambient excitation is a stochastic process. The response changes with structural damage. Jensen-Shannon Divergence(JSD) can be used to measure the difference between the random variable probability distribution. This dissertation is based on the structural damage characteristic quantity of time-domain data, combining with the relevant methods of nonlinear time series analysis. The new index of the testing structural damage is proposed based on the vibration-based structural damage detection technique. The main content is as follows.Research the theory of phase-space reconstruction using univariate time series. This paper will mainly analyze the method of the time delay determination and the embedding dimension select. The autocorrelation function method, modified false nearest neighbors algorithms are investigated to determine the two parameters, the two methods are well applied in the field of structural damage diagnosis.Analyze the basic principle of singular value decomposition, the realization of the algorithm is studied in detail. Verify the method is effective to restrain noise by simply supported beam model, the result show that: through noise reduction processing, the singular spectrum effective by noise can reflect the real characteristics of power system.We firstly introduce the concept of cross-entropy and the index of singular-spectrum cross-entropy, then the JS divergence theorem is deduced through the concept of information entropy. Jensen–Shannon divergence is used as an index for damage diagnosis.Through two different methods, the new index was tested on the benchmark model designed by ASCE task group. Reconstruct phase-space using the acceleration under different conditions, then through the singular value decomposition, the singular value is applied to the JS divergence. The influences of operational conditions were analyzed. Then we discuss the index of the damage location and compare the singular-spectrum cross-entropy with JS divergence. The result of which confirm the effectiveness of the proposed method. Verify the feasibility of JS divergence as a damage detection index.