A disjunctive belief rule-based expert system for bridge risk assessment with dynamic parameter optimization model

- Summarize advantages of utilizing DBRB expert system in modeling bridge risks.
- Provide more general analyses on the completeness of DBRB expert system.
- Propose a dynamic parameter optimization model for DBRB expert system.
- Propose an improved DE algorithm to search for global optimal parameter values.

Bridge risk assessment is an important approach to avoiding the safety accidents of bridges and ensuring the safety of the public. This can be done by investigating the relationship between bridge risks and bridge criteria. However, such relationship usually is highly complicated in actual situations. In this regard, many approaches were proposed to model bridge risks in the past decades. Particularly, four alternative approaches including the artificial neural network (ANN), evidential reasoning with learning (ERL), multiple regression analysis (MRA), and adaptive neuro-fuzzy inference system (ANFIS) were deeply analyzed and compared for bridge risk assessment. However, these approaches are restricted by their shortages. Thus, this paper utilizes the disjunctive belief rule-based (DBRB) expert system to model bridge risks, where the DBRB expert system is one type of the belief rule-based (BRB) expert system by considering disjunctive belief rules (DBRs) rather than conjunctive belief rules (CBRs) in a BRB. Furthermore, the dynamic parameter optimization model and improved differential evolution (IDE) algorithm are proposed to train the parameters of the DBRB expert system, where the model is applied to ensure the completeness of a DBRB and the algorithm is used to get the global optimal solution. For justification purpose, two existing parameter optimization models and nine alternative models developed by the ANN, ERL, MRA, and ANFIS are applied to assess bridge structures. Comparison results indicate that the DBRB expert system with the dynamic parameter optimization model is better than those alternative models and existing parameter optimization models.