PSI - Issue 62

Fabio Severino et al. / Procedia Structural Integrity 62 (2024) 276–284 Severino et al. / Structural Integrity Procedia 00 (2019) 000–000

280

5

● The entire life cycle of the AI system must be traced through blockchain technology, meaning every operation must be recorded within immutable ledger data structures, making them resistant to tampering ● Data confidentiality must be ensured by utilizing a blockchain that is private, permissioned, and supports data segregation. ● The data recorded in the private blockchain must be publicly notarized (without compromising confidentiality, as in the case of Hybrid DLTs), allowing for auditing by entities not within the private blockchain network (external audit). ● Each actor must sign its operations recorded in the blockchain using its own cryptographic key, making them non-repudiable. ● The rules and constraints of processes must be encoded as smart contract functions, ensuring their enforcement. ● The inferences produced by the AI system must be enhanced by issuing a certificate that identifies the data lineage, ensuring its provenance, reproducibility, and external auditability. In this use case, the first phase of the life cycle involves the collection of data that will form the training dataset. To train an AI model capable of recognizing structural defects from input images, the training data consists of a collection of example high-resolution images of bridges, taken by drones flown by a human operator, and annotated by human experts by marking the extent and the type of every defect visible, as shown in Fig. 1.

Fig. 1. image of a bridge with annotated defects.

Examples of the defects that can be identified in these images are: ● Crack (with horizontal, vertical, and oblique classification) ● Reinforcement (exposed, oxidized, corroded reinforcement) ● Humidity (categorized into light and dark stains to indicate passive or active humidity levels) ● Peeling ● Rust