Structural Health Monitoring: A Kriging Convolutional Network Approach for Time Series Imputation

The Structural Health Monitoring (SHM) and Non-Destructive Testing (NDT) are strategic experimental approaches for the conservation and safeguarding of historical architectonic heritage with complex soil conditions like the “Sassi” region of the Matera city in Italy. The presence of numerous underground structures and subterranean sites with aboveground masonry structures, even on multiple levels, affected by a set of “cavities” below floor level, because of strong and continuous, deep, and frenzied transformations of this part of the city, makes it indispensable to apply a monitoring and control system for the area [1]. The presence of these cavities poses a serious risk to the safety of the structures and the population of the city due to the potential collapse of buildings and structures. Nonetheless, today we increasingly see the implementation of Digital Twins (DT) [2], which, combined with the use of the Internet of Things (IoT), have determined a connection between the physical world and the digital one becoming increasingly mature and advanced, allowing much more comprehensive control over SHM systems [3, 4]. In this regard, “Data augmentation” is fundamental, since it arises from the concept of integration of the IoT and artificial intelligence (AI) paradigms, which, combined and integrated within systems and processes, can determine intelligent systems in order to use digital data much more efficiently and effectively. However, the spread of innovative 3D modelling techniques in Geographic Information System (GIS) environments, structured also on multiscale approaches for various Levels of Detail (LoD), or 3D GIS development environments that facilitate the management of multipath entities and point clouds, adopting evolved OGC standards or CityGML, is not accompanied by equally fast growth of IoT infrastructure. The basic idea is to make digital processes less dependent on human influence, which is often a limiting factor for the correct execution of processes, systematically applying AI and machine learning techniques to data acquired from IoT platforms, with the aim of "increasing" the informational power of these digital systems, including developing systems capable of anticipating or predicting certain phenomena. The ambition is to implement DT capable of updating in real time, adapting to data coming from geospatial sensors. Geospatial sensors, however, encompass a vast variety of devices that need to be able to detect numerous geographical and “non-geographical” information and characteristics of the real world.