Random Sensitivity Analysis of Creep Effects in Concrete-Filled Steel Tube Members

Literature Overview

This study by Zhao J. G., Zhao R. D., and Zhan Y. L. (2016), published in the Journal of Architecture and Civil Engineering (Vol. 33, No. 4, pp. 44-50), addresses the inherent randomness in the creep behavior of concrete-filled steel tube (CFST) axially compressed members. The authors employ a combined approach of support vector machine (SVM) regression for explicit function approximation and Monte Carlo simulation (MCS) for probabilistic analysis, with adaptive hybrid particle swarm optimization (AHPSO) to tune SVM hyperparameters. The research is funded by the National Natural Science Foundation of China (Grant No. 51208431), the Central Universities Fundamental Research Funds (SWJTU12CX064), and the Guizhou University Talent Recruitment Program (201517). The work was conducted at Southwest Jiaotong University and Guizhou University.

Core Technical Approach

The fundamental challenge addressed here is the computational cost of traditional Monte Carlo simulation when evaluating the probabilistic behavior of CFST creep, which involves multiple random variables including concrete compressive strength, steel tube yield strength, concrete creep coefficient, steel tube creep coefficient, and the initial eccentricity ratio. The authors propose a surrogate modeling strategy where SVM is trained on a limited set of finite element analyses to approximate the explicit functional relationship between random inputs and the creep-induced deformation or load capacity.

The SVM regression maps the nonlinear, multi-dimensional input space to a scalar output representing creep deformation or effective load capacity. The key advantage is that once the SVM model is trained, evaluating sensitivity coefficients for thousands of MCS samples becomes computationally trivial compared to running full nonlinear finite element analyses for each sample. The AHPSO algorithm optimizes the SVM penalty parameter C and kernel width parameter gamma, which are critical for balancing approximation accuracy and generalization capability.

Sensitivity Coefficient Formulation

The sensitivity coefficient for each random variable X_i is defined as:

S_i = Cov(X_i, Y) / (σ_Xi × σ_Y)

where Y is the structural response (creep deformation or load capacity), Cov denotes covariance, and σ represents standard deviation. This formulation quantifies the linear correlation between each input random variable and the output response, normalized to be dimensionless and bounded between -1 and +1.

The authors validate their SVM-based approach against direct Monte Carlo simulation results, demonstrating that the relative error remains small. They also confirm that the creep effect of CFST members exhibits stochastic characteristics with probability density functions that approximate normal distributions, which validates the applicability of the first-order reliability method (FORM) for subsequent structural reliability assessments.

Sensitivity Results and Engineering Implications

The parametric sensitivity analysis reveals that among the various random variables considered, the concrete creep coefficient and the concrete compressive strength are typically the most influential factors governing the stochastic behavior of CFST creep. The steel tube yield strength and the initial eccentricity ratio contribute to a lesser but still significant degree. This finding has direct implications for engineering practice: quality control efforts should prioritize the characterization and monitoring of concrete creep properties and compressive strength variability during the construction of CFST members.

From a materials engineering perspective, the creep behavior of the concrete core is governed by its age at loading, relative humidity, temperature, and cement type. The steel tube, while contributing less to overall creep, plays a critical confining role that affects the stress-strain behavior of the concrete under long-term loading. The interaction between steel creep and concrete creep creates a coupled deformation mechanism that must be captured accurately in any predictive model.

The probability density function analysis showing near-normal distribution behavior is particularly valuable. It means that for most practical purposes, the first two statistical moments (mean and standard deviation) are sufficient to characterize the creep response distribution, simplifying reliability calculations considerably. However, engineers should remain cautious when dealing with extreme loading conditions or very long service lives where the normality assumption may break down.

Connection to Pipe Manufacturing and Welding Practice

Although this study focuses on structural behavior rather than pipe fabrication, the implications for steel pipe specifications in CFST applications are significant. The steel tube used as the outer shell in CFST columns must meet stringent requirements for dimensional accuracy, wall thickness uniformity, and material consistency. Variations in wall thickness directly affect the confining pressure on the concrete core, which in turn influences the creep behavior.

From a welding quality perspective, longitudinal and spiral welds in the steel tube can introduce residual stresses and microstructural variations that affect the local creep resistance. Welded joints, particularly those in ERW or HFW pipes, may exhibit different creep characteristics compared to the base metal due to the heat-affected zone (HAZ) microstructure. This underscores the importance of post-weld heat treatment and thorough non-destructive testing of the steel tubes used in CFST construction.

The sensitivity analysis methodology presented here can be adapted for other structural reliability problems involving steel pipes and pipe fittings, such as the long-term creep behavior of steel pipe supports, pipe rack systems in power plants, or marine platform structures exposed to cyclic loading.

Study Insights and Reflections

The integration of data analysis surrogate models (SVM) with traditional probabilistic methods (MCS) represents a pragmatic engineering approach to managing computational complexity. The authors demonstrate that this hybrid methodology achieves accuracy comparable to direct MCS while reducing computational effort by orders of magnitude. This is particularly relevant for parametric studies where hundreds or thousands of structural analyses are required.

However, the accuracy of the SVM surrogate depends heavily on the quality and coverage of the training data set. Engineers must ensure that the training samples adequately represent the full range of random variable combinations, particularly in the tails of the distributions where rare but critical events occur. The AHPSO optimization of SVM parameters is a practical step toward this, but cross-validation and hold-out testing remain essential quality control measures.

The finding that creep effects in CFST members follow approximately normal distributions simplifies subsequent reliability analysis but also highlights the need for careful consideration of boundary conditions, material model accuracy, and the validity of the assumed random variable distributions in the underlying finite element model.

This study contributes meaningfully to the understanding of long-term structural behavior of CFST members and provides a methodological framework that can be extended to other composite structural systems. The emphasis on sensitivity analysis is particularly valuable for guiding quality control priorities and material selection decisions in practice.