Assessing the seismic residual capacity of damaged buildings after a major earthquake is critical to support decision-making regarding re-occupancy, repair, or demolition. This task typically requires collecting information on the observed earthquake-related damage to structural components in order to update their expected seismic response and perform additional safety and economic loss evaluations for the damaged structure. While an expert engineering judgment remains crucial to assess damage severity, modern machine-learning techniques such as Convolutional Neural Networks (CNNs) can represent a supporting tool to automate and standardise this process. From the earliest emergency phases, a CNN-based damage detection tool can be used to process large amounts of photographic documentation collected by non-experts or captured by devices such as cameras and drones. The information on the observed damage can be used to automatically update available analytical or numerical models of the structure - up to "Digital  Twin" models - returning a preliminary estimation of the seismic residual capacity of the damaged structure.

In this context, this paper presents and discusses a framework for post-earthquake seismic residual capacity assessment of damaged buildings that integrates a CNN-based damage detection tool. According to state-of-the-art procedures, the framework employs non-linear static (pushover) analyses - performed through either a simplified analytical/mechanical procedure or numerical (software-based) simulations -  and capacity reduction factors to modify the plastic hinge's behaviour of damaged structural components. The level of earthquake-related damage is automatically evaluated from images using a CNN based on a VGG16 architecture, trained on 5'000 RGB images sourced from existing databases and re-labelled by experts. The proposed framework is implemented for a case-study reinforced concrete structure for illustrative purposes. The seismic performance before and after the damaging earthquake is evaluated in terms of a capacity-to-demand safety index and expected annual losses. The results highlight the potential of the CNN-based framework to support emergency planning and decision-making, particularly for large building portfolios.