Overview / Usage

Structural Health Assessment (SHA) is a nondestructive technique to monitor the functionality of a certain structure or system. It utilizes the measured dynamic response of the system to detect, localize and quantify any damage or failure in that system. Current parametric approaches try to predict the damage in the structure by identifying its stiffness parameters. Such approaches require complicated analytical modeling for the structure, as well as complicated analytical analysis to extract the system parameters.
Statistical modeling is currently trending, it uses a library of signatures for the structure responses, any further response from the structure can be classified as one or a combination of the damage scenarios recorded in that library.
This current proposal aims to utilize a pre-generated library of damage scenarios to classify any new damage that occurs further. The investigated structure is a two-story single-bay frame structure as shown in the figure.

Questions to be answered using this analysis:
Having the excitation and response of the structure, what is the damage scenario that best describes the current state of this structure? That is, is it a single or multiple element damages? What are those damaged elements? And to what extent is each damage?

Data analytics technique will be used:
This problem does not require clustering, it is only a classification problem. The main challenge is the nonlinearity. Having the records of two structure responses for two single-element damage scenarios, they cannot be both combined to predict the response of a multiple-element damage. Thus, we need to consider the nonlinear effect in the classification.
An initial approach is the Nearest Neighbour, it is simple but cannot account for nonlinearity. This requires a huge number of scenarios to be generated and then fed into the feature space for further analysis. A further approach to be applied is LASSO that could count for nonlinearity and cross relations between different elements of the structure. Other approaches can be used later based on the achieved progress in the previous two approaches.

Collaborators