PSI - Issue 64

Shirley J. Dyke et al. / Procedia Structural Integrity 64 (2024) 21–28 Dyke et al / Structural Integrity Procedia 00 (2019) 000–000

26 6

leveraging engineering drawings stored in digital formats (2023). Dimensions of the walls and columns are extracted from structural drawings to compute the Hassan Index, a well-known seismic vulnerability index for reinforced concrete buildings. By employing sophisticated computer vision and deep learning algorithms, the researchers streamline the process to enable rapid regional assessments of building inventories.

Fig. 3. Image classification tools implemented in ARIO.

5. Risk-informed Policy and Decision-support Using AI Despite significant strides in technology, the inherent limitations of current AI methods introduce risk, particularly when applied to critical sectors such as infrastructure safety and community resilience. Lenjani et al., shed light on this challenge by developing a scalable method designed to prioritize building inspections with the dual goals of minimizing expected losses and accelerating recovery efforts (2020). Fragility functions that estimate the likelihood of reaching various damage states based on seismic intensity guide the prioritization of inspections. The methodology seeks to reduce the expected loss to the community, factoring in both direct inspection costs and the potential repercussions of misclassifications in the AI methods to empower managers to determine a budget to prepare for post hazard inspections based on their risk tolerance. This work highlights the role of managing limited resources during post-disaster recovery. Earlier we discussed the automated use of the bridge inspection database to empower engineers to perform seismic vulnerability assessment for retrofit prioritization by classifying bridge substructure types (Zhang, et al. 2023a). Classification results corresponding to high confidence and low risk would be automatically entered in the database. However, results corresponding to bridge types with higher risk would be selected for visual confirmation. It would be logical to link the automated assignment of the substructure types to the confidence with which the substructure classification outcomes are determined from the images. The work also went a step further to develop a risk-tolerance based method to set a risk-informed budget for this task. Several sample results for different regions are shown in Figure 4. This work enables a rapid seismic vulnerability assessment to be conducted more frequently by the asset manager to deal with an evolving inventory and changing policies. Zhang et al. further explore the application of AI in the domain of bridge inspection and management (2023b). Reinforcement learning (RL) is used as a method to optimize bridge inspections. This method allows for customized inspection schedule based on a bridge’s location, exposure to hazards, and past evidence of deterioration. Simulation results that consider degradation of the deck due to chloride-induced corrosion are used to demonstrate the method here. Utilizing a CNN as the agent within the RL system, the method processes the current state of the bridge, including