Ontology-Based Approach to Risk-Factor Identification to Support the Management of Provisions in Bridge Design

Abstract

Requirement management is crucial in the design stage of infrastructure project delivery to ensure the designed facility satisfies the owner expectations and avoids any costly redesign, rework, and disputes. The requirements are primarily described in design codes and standards. The relevant requirements from codes are transformed into a set of in computer-readable rules required for computer-aided design and compliance verification. Prior to the construction of rules, the requirements in codes must be converted into an organized, structured format in terms of risks to support easy retrieval of the relevant applicable requirements. However, the unavailability of risk information in the requirement text makes it challenging for a designer to extract the relevant requirements of specific risks to support design tasks, including compliance checking. The manual process of requirement classification according to risk factors is time-consuming, laborious, and error-prone because the codes are mostly voluminous, including thousands of specifications. Much less attention has been paid to develop an automated framework to organize the requirements in terms of risks addressed in them. To address this need, this study has attempted to develop an ontology-based framework to identify relevant risk factors addressed in bridge design requirements to support requirement management. The nine risk factors of bridge engineering used in this study included flood, earthquake, fire, snowfall, wind, temperature, overloading, vessel collision, and blast loading. A risk ontology was developed to represent the conceptualized semantic knowledge associated with each of the nine risk factors. The algorithm was validated on a human-annotated requirement dataset based on the AASHTO bridge design specifications and state design manuals. The developed model yielded a Spearman, Kendall tau, and Pearson correlation coefficient of 0.7223, 0.6021, and 0.7222, respectively. The proposed model is expected to improve specifications retrieval for rules construction, thereby enabling ease of requirement classification for the design process.