Issue 46

F. Bazzucchi et alii, Frattura ed Integrità Strutturale, 46 (2018) 400-421; DOI: 10.3221/IGF-ESIS.46.37 415 system that is able make emerge the damage from an inspection could be very useful in assessing the safety conditions. Moreover, it allows a previous screening for where to install detailed SHM tools and help for their result interpretation. A promising way to pursue is represented by portable Digital Radiography (DR) devices [32]. In contrast to film or Computed Radiography, (DR) uses an X-ray digital image capturing device (Figure 20a). X-rays allow an "under skin" reading of the material, allowing to establish the presence, the position and, with a good approximation, the diameter of the reinforcements, the cables, the strands. Moreover, if the protocol to be used is suitably calibrated, the x-rays allow to detect anomalies such as the presence of voids, the presence of non-coherent materials and segregation effects. The system is usually composed by an emitter, a real time display and a shield detached detector (Fig. 20(a)). This technology enables to generate immediate and high-quality X-ray images while keeping radiation levels to the minimum. Fig. 20(b) shows the results of an X-Ray test conducted in Politecnico di Torino laboratory Fig. 20(c). The system exhibited a significant power of penetration, completely scanning the 150x150x150 mm concrete cube. Furthermore, the system was able to detect additional reinforcement positioned beyond the cube, then has a great potential also for the inaccessible portion of a structure (Fig. 20(d), Fig. 20(e)). The all system weight less than 200 kg (the heavier component is the detector, 150 kg in its largest dimension) and can be easily carried by a robotic-arm installed on a vehicle (Fig. 21). Inspection of I-beams could be carried by a scraped scanning. Of course, this method represents a screening procedure but could be extremely reliable and effective for a preliminary census or fast track assessment. Figure 21 : Vehicle screen scanning. Artificial Intelligence: Computer Vision and Pattern Recognition Artificial intelligence (AI) is a definition that groups altogether the techniques in computer science that solve a problem (or a task) by building an algorithm-based (or a series) agent that acts in a certain environment and not by following a series of dictated operations [33]. From 1965, the scenario of AI broadened so vastly that each its subfield could be considered as an independent field of study, as for Machine and Deep Learning (DL), Computer Vision (CV), Soft Computing, Human-inspired Intelligence, and Neural Networks (NN). To exploit the process of an agent to act, the ability of computational platforms to “learn” has to be explored. In general, when a problem is ascribable as a goal- function for completing the task, we speak of classic Machine Learning. The agent is an algorithm that solves the problem by maximizing its goal completion in the fastest way. Each ameliorative step it is transformed in boundaries for the algorithms series and it is defined learning process. These series are usually neural networks, that operate with different goals in every node (neuron) of the net [34]. The advantages of problem solving with classic AI are mostly (i) the desired task is completed without human intervention and (ii) the system is self-improving. The revolution of big-data changed, and boosted, the practice of AI. If the system is well trained, it could detect, classify, recognize and predict results in particular scenarios, as human brain does in everyday life. In imitating human intelligence and perception, the most advanced field is Vision. The reason is connected to the large similarity between the structure of the human eye and