PSI - Issue 62

Lorenzo Principi et al. / Procedia Structural Integrity 62 (2024) 89–96 Principi L./ Structural Integrity Procedia 00 (2019) 000 – 000

91

3

2. Dataset The intent of this work is to use a reduced set of information, which needs to be easily and quickly recovered exclusively from documents, thereby avoiding surveys or inspections. Data is derived from the L0 Reports which are compiled during the mandatory census activities, therefore this information is generally available. The dataset collected information of 423 highway bridges within the Italian territory. The LoD and the Structural and Foundational Attention Class are outputs of Level 2 Reports and are intended to be the prediction targets of the model. Filling L0 Reports requires a significant amount of information, and Agencies often miss some data. Therefore, there is a necessity to choose a reduced, informative, and readily accessible set of information. Thus, this work suggests the use of a smaller subset of information, referred to as Input Features (IF). The chosen parameters need to be adequately explanatory for the generalization of the model and quickly available. According to some considerations reported in Assaad and El-Adaway (2020a,b), Xia et al. (2021), and Alogdianakis et al. (2022), a subset of IF is chosen among data contained in L0 Reports, considering those previously demonstrated to be effective in estimating bridge LoD. Moreover, IF are selected to be general and immediately available by Agencies. Errore. L'origine riferimento non è stata trovata. provides a comprehensive list of the IF, each accompanied by a brief description. Selected IF, as presented in Table 2, are diverse in nature and are described from a statistical point of view as indicated in Piccolo (2020). The description in the table includes not only statistical aspects but also additional details such as possible values (specific of the considered case study), ranges, and classes.

Table 1. Selected Features and Descriptions

Selected Features Max span length

Risk Parameter Classification

Features Description

Exposure Exposure Exposure Exposure

Indicates the length of the longest span Indicates the average daily volume of traffic

ADT

Obstacle type Alternatives

Typology of the overcome obstacle

Presence of other roads to service bridge traffic Indicates the average daily trucks volume of traffic Maximum allowable mass on the bridge Indicates the length of the structure Indicates the number of spans of the structure Indicates the static scheme of the bridge

ADTT

Hazard Hazard

Load limit Total length

Vulnerability Vulnerability Vulnerability Vulnerability Vulnerability Vulnerability

Number of spans

Static scheme Deck material

Indicates the deck material typology

Bridge Age class Design Code class

Range in which the date of construction of the bridge falls into *

Class of design code *

*Age classes are Class A, Class B and Class C

3. Data preprocessing In order to not reduce the predicting performance of the model (Asaithambi 2017), each IF selected needs a different pre-processing strategy according to its nature. As can be seen in Table 2 for numerical features, i.e. Total Length, Max span length, and Number of span s , they show different magnitudes. Hence, scaling is needed to bound these values within a consistent range. To do this, an ad-hoc scaling function is proposed, which is a modified version of an exponential decay function, represented by ( ) = 1 − − (1) with being a real positive number selected case by case, for each numerial feature. Each equation is obtained by setting a threshold value, called , such that ( )=0.90 . This value delineates a transition between two behaviours. For values below the threshold, the function exhibits a rapid growth rate.