Open Access

ABSTRACT

Influence of the Regression Error of the Response Surface to the Diagnostic Accuracy of the Unsupervised Statistical Damage Diagnostic Method

A.Iwasaki¹, K.Yuguchi², A.Todoroki³, Y.Shimamura⁴

1 Department of mechanical engineering, Gunma university, 1-5-1 Tenjincho, Kiryu, Gunma, Japan
2 Grad. school of engineering, Gunma university, Japan (Thermos K.K. at present)
3 Dept. of Mech. Sciences and engineering., Tokyo Tech, 2-12-1 Ookayama, Meguro-ku, Tokyo, Japan
4 Dept. of Mechanical Engineering, Shizuoka university, 3-5-1 Johoku, Hamamatsu, Shizuoka, Japan

The International Conference on Computational & Experimental Engineering and Sciences 2008, 6(3), 183-188. https://doi.org/10.3970/icces.2008.006.183

Abstract

The present study is about study on the diagnostic accuracy of the unsupervised damage diagnosis method named SI-F method. For the health monitoring of existing structures, modeling of entire structure or obtaining data sets after creating damage for training is almost impossible. This raises significant demand for development of a low-cost diagnostic method that does not require modeling of entire structure or data on damaged structure. Therefore, the present study proposes a low-cost unsupervised statistical diagnostic method for structural damage detection. The proposed method statistically diagnoses structural condition by means of investigating the change of a response surface which conducts the system identification between sensor outputs. The response surface is calculated as a regression model of relationship between multiple sensors. The shape of the response surface is changed reflecting the change of the structural condition. In this method, the change of the response surface is statistically investigated with the F-test. In the F-test, the threshold of normal or damaged condition is decided with only theoretical F-probability distribution. This theoretical F-distribution is easily calculated using the response surface parameters. Therefore, diagnosis is conducted by means of only intact data used for the reference data. This means the proposed method doesn't require information about the damaged condition. \newline Since the SI-F method is able to detect the damage in the structure by judging the deviation from the normal state, it is important to reduce the false positive detection for raising the reliability of the structure. In the present study, to clarify the relationship between the condition of the false positive detection, diagnostic accuracy and the regression error of the response surface, several numerical simulations were carried out.

---