Identification of damaged member in a truss structure using acoustic emission technique aided by SVM

Abstract

Structures are prone to damage, and detecting them at their very initiation is extremely important for taking corrective measures. Truss is a very important civil engineering structure having a complex geometry: width and thickness being very small compared to the length of a member and the presence of discontinuities in the form of joints. Most of the existing techniques identify damages after they have grown to a substantial degree. For early detection of damages, the acoustic emission (AE) technique may be used effectively as the initiation of damage itself results in the emission of AE wave, analysis of which may lead to detection of the damage. Additionally, signature behavior of the virgin structure is not necessary, unlike many other nondestructive testing methods. Existing studies largely use the time of arrival (TOA) of AE wave at the sensor location(s) in the formulation for damage localization. However, TOA is significantly affected by reflected waves from the edges of truss members and attenuation of signal strength during its passage through joints. In addition, TOA suffers from a lack of objectivity in prescribing the relevant threshold value. Accordingly, damage localization may be affected. In this context, a machine learning approach is very promising if appropriate signal features are used for training. Accordingly, in the present study, an effort is made to develop a support vector machine model for the localization of a truss member, where damage has been initiated. The model is trained and then tested using different sets of experimental data, collected from a laboratory-scale truss. The results of the localization, as predicted by the model, are found to be encouraging when compared with the physical observation.