Towards Multiclass Damage Detection and Localization using Limited Vibration Measurements

Traditional vibration-based damage detection methods provide structural health information based on their measured data (i.e., acceleration and displacement response). Over the last few decades, various model-based and time-frequency methods have shown great promises for damage identification and localization. However, the existing methods are unable to perform satisfactorily in many situations, including the presence of limited sensor measurements and training data, detection of minor and progressive damage, and identification of multiclass damage, creating constraints to make them free of user-intervention and implemented using the modern sensors. The main objective of this thesis is to develop algorithms capable of damage identification and localization using limited measurements that can address the limitation of the traditional methods while providing a minimal to no user-intervention damage identification process. The proposed research in this thesis involves casting damage detection problems as non-parametric and autonomous with the least user intervention. Progressive damage identification is presented using novel time-frequency methods, such as synchrosqueezing transform and multivariate empirical mode decomposition, showing improved sensitivity of identifying minor damage over traditional methods. A basis-free method, such as multivariate empirical mode decomposition, is employed for damage localization using limited sensors. The acquired vibration measurement is decomposed into its mono components, and a damage localization index based on modal energy is proposed to overcome the need for a large number of sensors. The limited measurement aspect of damage localization is explored by selecting fewer sensors, and it is shown that with limited measurements, the proposed method is as effective as a total number of measurements equals the number of degrees of freedom of the model. To create an autonomous damage identification framework, Artificial Intelligence-based methods are explored the first time for multiclass damage classification and localization. Due to the lack of availability of a large amount of data, the acquired vibration data is augmented using windowing of the data per damage class. A novel window-based one-dimensional convolutional neural network is explored to classify sequential time-series of vibration measurements with only one hidden layer. The robustness of the proposed method is further evaluated by a suite of parametric and sensitivity analysis. Improvement of this method is further accomplished by implementing a windowed Long Short-term Memory network capable of learning long-term dependencies of the sequential data. Finally, the proposed methods are validated using a suite of experimental and full-scale studies, including a high-rate dynamics experimental testbed, a stadia prototype experimental setup, the MIT green building, and the Z24 bridge.

Name thesis:degree_name

Ph D

Discipline thesis:degree_discipline

Civil and Environmental Engineering

Grantor dc:publisher

The University of Western Ontario

Year dc:date.issued

2021

Author dc:creator

• Sony, Sandeep

Advisor dc:contributor.advisor

• Sadhu, A

• Deep Learning
• Damage Detection
• Vibration
• SHM
• Limited sensors

Language dc:language.iso

en_ca

Handle dc:identifier.uri

https://hdl.handle.net/20.500.14721/30889

OAI identifier oai:identifier

oai:uwo.scholaris.ca:20.500.14721/30889