Reliability Assessment of Large-Span CFST Arch Bridges Using Hybrid Algorithms

Literature Overview

This paper by Cui Fengkun et al. (2016), published in the Journal of Xi'an University of Architecture and Technology, addresses the serviceability reliability assessment of large-span concrete-filled steel tube (CFST) arch bridges. The study is grounded in the Chinese design code for highway CFST arch bridges (JTG/T B07-01) and the technical specification for CFST arch bridges (JGJ/T 384). The authors construct a systematic reliability evaluation model and propose a hybrid algorithm combining backpropagation (BP) neural networks, uniform design methodology, and an improved Hasofer-Lind-Rackwitz-Fiessler (HLRF-JC) method to handle the implicit and highly nonlinear limit state functions inherent to such complex structures. The work was supported by the National Natural Science Foundation of China (Grant No. 50808019) and the China Scholarship Council (Grant No. 201606560011).

Core Technical Approach

The fundamental challenge in reliability analysis of large-span CFST arch bridges lies in the implicit nature of the limit state function. In conventional reliability methods, the limit state function g(X) is often unavailable in explicit analytical form, requiring numerical or experimental solutions for each function evaluation. This makes traditional first- and second-order reliability methods (FORM/SORM) computationally expensive and sometimes impractical for complex bridge structures.

The hybrid algorithm proposed in this paper integrates three complementary techniques:

Technique	Role in the Hybrid Algorithm	Key Advantage
BP Neural Network	Surrogate model approximating the implicit limit state function	Reduces repeated finite element analyses
Uniform Design Method	Efficient sampling of the input parameter space	Minimizes sample size while maximizing information content
Improved JC Method	First-order second-moment (FOSM) reliability index computation	Handles non-normal random variables via equivalent normal transformation

The BP neural network is trained using samples generated by the uniform design method, which distributes experimental points more uniformly across the parameter space than traditional random sampling. The trained neural network then serves as a surrogate for the limit state function, enabling rapid reliability index calculation through the improved JC algorithm. This approach significantly reduces computational cost while maintaining acceptable accuracy.

Interpretation of Key Findings

The numerical examples and actual bridge verification cases reveal several important observations:

The reliability indices for main arch stress and deflection under serviceability limit state (SLS) conditions significantly exceed the code-specified target values, indicating substantial optimization potential in the main arch design.
The reliability indices for transverse beams and columns are comparatively lower and warrant increased design attention.
The hybrid algorithm demonstrates feasibility and accuracy for reliability analysis of complex CFST arch bridge structures with implicit and nonlinear limit state functions.

From a practical engineering perspective, this finding suggests that the current design codes may be overly conservative for the main arch members of large-span CFST bridges, while the secondary structural components (transverse beams, columns) represent the critical reliability links in the overall system. Engineers should consider redistributing material from the main arch to strengthen secondary members, thereby achieving a more uniform and efficient reliability distribution across the structure.

Integration with Engineering Practice

In my experience with steel pipe and structural engineering, the reliability assessment methodology presented here has direct implications for the design and quality control of CFST bridge components. The main arch ribs of large-span CFST bridges are typically fabricated from high-strength steel tubes (such as Q345qD or Q420qD grade steel per GB/T 22510) filled with high-performance concrete (C50 or above). The fabrication quality of these steel tubes—including dimensional tolerance, wall thickness uniformity, and weld integrity—directly influences the actual reliability index of the completed structure.

The finding that secondary components have lower reliability indices highlights the importance of proper quality control for transverse beam and column fabrication. These components often use lower-grade steel tubes (e.g., Q235 or Q345B) with thinner walls, and their welding connections are more susceptible to fabrication defects such as incomplete fusion, lack of penetration, and weld undercut. Engineers should ensure that welding procedures for these critical members comply with relevant standards such as GB 50661 (Steel Structure Welding Code) and that non-destructive testing (NDT) coverage is adequate.

Key Questions and Reflections

Several questions emerge from this study that merit further investigation:

How does the reliability index of the main arch change when the steel tube material properties (yield strength, elastic modulus) are considered as random variables with realistic statistical distributions derived from actual material test data, rather than assumed normal distributions?
The paper focuses on serviceability limit state; what is the reliability performance under ultimate limit state (ULS) conditions, particularly considering the progressive collapse risk of CFST arch bridges?
How do fabrication defects in the steel tubes (wall thickness variations, ovality, weld residual stresses) affect the actual reliability index compared to the theoretical values presented?

These questions connect the theoretical reliability framework to practical fabrication and quality assurance concerns, which are central to my professional expertise in steel pipe manufacturing and welding.

Study Insights and Implications

The hybrid algorithm approach presented in this paper represents a significant methodological advancement for structural reliability analysis of complex CFST bridge systems. The integration of neural network surrogate modeling with efficient sampling and reliability index computation provides a practical solution to the computational bottleneck of implicit limit state functions. For engineers involved in the design, fabrication, and quality control of CFST bridge components, the key takeaway is that the main arch members are generally over-designed in terms of serviceability reliability, while secondary components represent the true reliability bottlenecks. This insight should guide both design optimization and quality assurance resource allocation in future projects.