Description
Bridges worldwide are failing at increasing rates as they heavily suffer from deterioration, leading to the transport infrastructure system disruptions and millions in losses. For example, the failure of the Polcevera bridge alone cost € 359 million in the immediate wake, with estimated annual losses to the Italian economy in the vicinity of one billion euros. Despite the aid of technologies involving digital measurements and data collection from remote and structural health monitoring, bridges still fail. The reasons behind those failures are mainly due to the fact that damage is not easily discoverable despite the data at hand and the scarce availability of design drawings to the asset assessors, especially for old bridges. Furthermore, the construction sequence of the asset or level of prestressing are often unclear, hence the as-built condition and properties of bridges are largely unknown. Deck scanning techniques are often expensive therefore are rarely adopted by transportation stakeholders and operators. This is a capability gap that can only be filled with engineering judgement as well as meaningful data collection and evidence in aid of decision-making. This paper provides a unique approach for undertaking a preliminary bridge damage assessment. The landmark balanced cantilever bridge of Polyfytos, located in the North-West part of Greece, was utilised in this paper. Heterogenous data and evidence were initially collected and deployed in order to inform and backtest an advanced numerical model of the bridge aiming to match measured deck displacements. Damage scenarios were then examined to explain vertical deflections accounting for a range of concrete Young’s Modulus and variable damage patterns in the tendons. The unsupervised machine learning algorithm k-Nearest Neighbours (k-NN) was finally adopted to strengthen the damage identification capability and classify the observed bridge deflected shape−for which the Young’s modulus has been identified−to the closest training set sample damage scenario. It was showcased that the presented methodology could be utilised to efficiently interpret deflections that are measured on the bridge deck.
