Abstract
A variety of civil infrastructure assets such as bridges, pipes and railways form an
integral part of modern societies. However, these structures are vulnerable to changes
in environmental conditions and physical or direct damages. These vulnerabilities have
brought rise to Structural Health Monitoring (SHM) systems, which are installed in civil
infrastructure assets to monitor the health of structures through installed sensors. SHM
is achieved by implementing techniques that identify, localise and assess the damage
on infrastructure assets.
Structural elements in metallic infrastructures assets have been connected using rivet
joints and bolts since 1900s. The integrity of these connections is a crucial factor in the
overall stiffness and strength of a structure; hence, it is beneficial to install a damage
identification system which monitors the dynamic response within connections and
detects any differences that may arise due to changes in connection characteristics.
Previous studies investigated the dynamic response through modal-based properties
where modal damping is one of the least researched topics due to mathematical
complexities in obtaining damping matrix and limitations in the traditional methods to
obtain damping ratio in both time and frequency domains. Probability Distribution Decay
Rate (PDDR) algorithm has been proposed which seems to be able to overcome the
limitations in time domain to detect changes in the overall damping by observing
changes in the statistical parameters. However, PDDR method limitations are following:
(1) was tested on only sensors that is placed close to structural connection with
loosened bolts; (2) only achieves levels 1 and 3 of Rytter’s damage classification
(detection and quantification).
Several techniques such as, Data fusion, damage localisation, supervised and
unsupervised learning, and dimensionality reduction technique were implemented to
PDDR algorithm to fuse the distribution data together and observe any deviation in the
physical condition of the structure and localise the damage in the structure with bolted
connections. Comparison was done between Kalman and Bayesian fusion
methodologies using single story frame and 4-Storey steel frame datasets and it
showed an improvement in detection, localisation and classification from individual
sensors.