Posters

The research’s main contribution is the examination of tools and methods for the application of structure-from-motion, photogrammetry, and Heritage/Historical Building Information Modelling (H-BIM), while understanding its flaws and proposing a plan for what the definition and structure of the field concerns. The project’s purpose, therefore, is not just the mere study and application of 3D modeling production, but also to define H-BIM criteria and application methods during the process of creation of a piece, from its execution to its subsequent maintenance. The research aims to investigate, through an experimental development process, the range of potential applications of this methodology, contemplating the use of three-dimensional modeling combined with H-BIM. The field of application concerns every artwork characterized by a high technological content or significant interconnection among architectural and structural aspects and, particularly, when the intervention needs coordination and collaboration among heterogeneous disciplines whose integration is considered fundamental. The case study presented concerns the most significant building University of California San Diego campus: the Geisel Library, a brutalist architectural masterpiece by architect William Pereira.

The Terrestrial Water Storage (TWS) is a key variable of the water cycle, as it includes all water reservoirs over the land and in the subsurface. Changes in the level of the global freshwater storage (Terrestrial Water Storage Anomalies, TWSA) may be generated by natural variability, climate change, and human activities. The Gravity Recovery and Climate Experiment (GRACE) mission provides estimates of the TWSA at the global scale with high accuracy. However, GRACE data present gaps and do not go back further than 2002. Moreover, the influence of climate variability and human pressure on the global water storage is not fully understood. Here, we present two machine learning models, i.e., a Long Short-Term Memory (LSTM) and a Bidirectional LSTM, that are able to reproduce the monthly time series of TWSA at the global scale from meteorological forcings (climate-driven models). Our results show that our models consistently reconstruct TWS fluctuations prior the GRACE era, fill the gap in the GRACE dataset, and assess the influence of climate variability on the observed TWSA globally.

Due to the lack of familiarity and trust among professionals, the use of fully automated procedures in structural health monitoring (SHM) is still limited in civil engineering. While visual inspections remain the most common method to assess structural components, electrical impedance tomography (EIT) is becoming popular for identifying strain states and disconnections in conductive elements. EIT involves inferring the electrical impedance of a conductive body from measurements collected using electrodes to form a tomographic image of the body. Recently, a new technology has been proposed, which consists of spraying conductive material onto structural elements and interrogating it periodically through electric pulses to identify cracks. However, EIT solutions obtained by solving a weak formulation of Kirchoff’s law (an elliptic partial differential equation) recursively, are sensitive to noise and can deteriorate over time due to material aging. To improve crack detection and localization performance, this study employs brain-inspired deep neural networks trained using simulated datasets and transfer learning, which allows for incorporating crack annotations obtained from visual inspections on similar structures. The results demonstrate that this approach outperforms traditional methods for identifying cracks in complex damage patterns.