Latent variable models can be used to eliminate the environmental or operational effects from the data without measuring the underlying variables and resulting in an increased reliability of damage detection. A method is proposed, which also utilizes the available environmental or operational variables. The method is based on the missing data analysis, in which each feature is estimated in turn using the other features and also the available environmental or operational variables. As damage detection is solely based on the measuremets, training data from the undamaged structure under different environmental or operational conditions are needed. Compared to many other latent variable models, the main advantage of the proposed method is that there are no parameters to be adjusted. The main disadvantage is a higher run time. The method is verified in a numerical study of a vehicle crane with a varying configuration and in an experimental study of a bridge structure under environmental variations. All damage cases were detected using the proposed approach, whereas no indications of damage resulted using the features directly. The importance of the measured environmental or operational variables for damage detection was found to be low, because the features typically consisted all the relevant information.
[1]
Harold W. Sorenson,et al.
Parameter estimation: Principles and problems
,
1980
.
[2]
Bart Peeters,et al.
System identification and damage detection in civil engineering
,
2000
.
[3]
Gaëtan Kerschen,et al.
Structural damage diagnosis under varying environmental conditions—Part I: A linear analysis
,
2005
.
[4]
Douglas C. Montgomery,et al.
Introduction to Statistical Quality Control
,
1986
.
[5]
P. Guillaume,et al.
A robust singular value decomposition for damage detection under changing operating conditions and structural uncertainties
,
2005
.
[6]
Hoon Sohn,et al.
Statistical Damage Classification Under Changing Environmental and Operational Conditions
,
2002
.