Structural damage quantification using ensemble‐based extremely randomised trees and impulse response functions

This paper presents a development and application of decision tree-based ensemble technique, extremely randomised tree (ERT) as a multi-output regression model in structural damage quantification of civil engineering structures. Acceleration responses are measured from structures when an impact force is applied. Impulse response functions as structural vibration properties are extracted from the acceleration responses and are processed as input to the ERT. Moving averaging with a suitable window size is performed to reduce the effect of noise, and principal component analysis is performed further for the dimensionality reduction. The damage level is defined in terms of elemental stiffness reduction. Both numerical and experimental studies are conducted to investigate the capability of using the proposed approach for structural damage identification and quantification. The numerical studies are carried out on a simply supported beam and experimental validations on a steel frame structure in the laboratory. From the results, the proposed method can provide good elemental structural damage quantification results. The computational time for the proposed approach is much less than random forest (RF) technique that has been used for the same application using acceleration responses. The performance of the proposed approach is compared with RF in terms of identification accuracy and training efficiency.