Second-order cone programming for distributionally robust compliance optimization of trusses considering input distribution uncertainty

Takumi Fujiyama, Yoshihiro Kanno · 2025

Read Second-order cone programming for distributionally robust compliance optimization of trusses considering input distribution uncertainty by Takumi Fujiyama, Yoshihiro Kanno (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-01678.txt.

Introduction

Takumi Fujiyama and Yoshihiro Kanno design trusses under distribution doubt in two thousand twenty-five. Loads on a truss are unsure, and past data on them are scarce. Scarce data can spoil designs tuned to a guessed distribution. Past confidence based methods strain on cost and accuracy for hard constraints.

Their method pairs two goals for the worst case of compliance. One is the mean, the other a tail risk gauge. Kernel density estimates stand in for a fitted distribution. The problem cuts down to a second order cone program for two kernel types. Tests trace the Pareto front of the two goals.

Catalog record

CREATORTakumi Fujiyama, Yoshihiro Kanno
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE76 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-01678

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Second-order cone programming for distributionally robust compliance optimization of trusses considering input distribution uncertainty Takumi Fujiyama, Yoshihiro Kanno Abstract Reliability-based design optimization (RBDO) is a methodology for designing systems and components under the consideration of probabilistic uncertainty. In practical engineering, the number of input data is often limited, which can damage the validity of the optimal results obtained by RBDO. Confidence-based design optimization (CBDO) has been proposed to account for the uncertainty of the input distribution. However, this approach faces challenges, computational cost and accuracy when dealing with highly nonlinear performance constraints. In this paper, we consider the compliance minimization problem of truss structures with uncertain external forces. Armed with the advanced risk measure, conditional Value-at-Risk (CVaR), we formulate a bi-objective optimization problem for the worst-case expected value and the worst-case CVaR of compliance, which allows us to account for the tail risk of performance functions not addressed in CBDO. Employing kernel density estimation for estimation of the input distribution allows us to eliminate the need for modeling the input distribution. We show that this problem reduces to a second-order cone programming when assigning either uniform kernel or triangular kernel. Finally, through numerical experiments, we obtain the Pareto front for the bi-objective optimization problem of the worst-case expected value and CVaR of compliance of truss structures, and confirm the changes in the Pareto solutions. ∎ Journal: Optimization and Engineering MSC 74P10: 90C15 90C90 90C31 1 Introduction 1.1 Background Design optimization of structures is a methodology to obtain a reasonable design of a structure by using a mathematical model representing the structural behavior and solving an optimization problem based on the model. The design domain and input data such as the external forces are usually given, and the optimal design is often obtained by maximizing a performance function of a structure under the constraint of an upper bound on the volume of the structure. Structures in the real-world are subjected to various uncertainties, such as external forces that cannot be anticipated during the design phase and variations in material properties that occur during the manufacturing process. Structures designed using deterministic models can possibly be vulnerable to these uncertainties in input data, which may result in compromised safety and performance in real-world environments. Therefore, it is pivotal to consider the uncertainties in input data during the design phase and perform optimization to ensure the safety and high-performance of designed structures. Reliability-based design optimization (RBDO) [ 1 , 16 ] is one of the probabilistic methods that considers such uncertainty in design optimization of structures. In RBDO, uncertain parameters are treated as random variables that follow a certain probability distribution. Since the values of the performance function of structures can be regarded as random variables, uncertainties are considered by imposing constraints on the reliability, that is, the probability that the values of the performance function satisfy the constraint conditions. Recent studies have proposed scenario optimization approaches to RBDO that address the challenges of probabilistic constraint satisfaction in the presence of epistemic uncertainty. For example, Rocchetta et al. [ 37 ] introduced a framework based on sampling and scenario-based formulation to ensure reliability with finite samples. An extension of this work was later presented in [ 38 ] , where the authors proposed a soft-constrained modulation of failure probability bounds to handle risk and reliability in a unified manner. Recently, there has been significant progress in the study of RBDO that also considers the uncertainty in the probability distributions of uncertain parameters, called confidence-based design optimization (CBDO) [ 33 , 22 , 43 ] . When sufficient input data is not available, it is difficult to accurately estimate the input probability distribution, making it challenging to properly evaluate the structural reliability. As a result, designs obtained through RBDO can become unsafe. In CBDO, the uncertainty in the probability distribution is considered by imposing constraints on the confidence, which is the probability that the reliability constraints are satisfied. Moon et al. [ 33 ] proposed a method based on a hierarchical Bayesian model and the Monte Carlo simulation (MCS). This method features a double-loop MCS, where the inner loop is for the reliability assessment and the outer loop is for the confidence assessment. This double-loop structure results in a very high computational cost. Additionally, it assumes a specific type of input distribution, which can lead to an unreliable design if the assumed distribution type differs from the true distribution. Jung et al. [ 22 ] proposed a method that replaces the MCS-based reliability assessment with the first-order reliability method (FORM). FORM approximates the reliability by linearizing the performance function, which can reduce accuracy for a highly nonlinear performance function. This method still retains a double-loop MCS structure with the outer loop for the confidence assessment, and thereby maintains high computational cost. Wang et al. [ 43 ] proposed a single-loop method using the second-order reliability method (SORM) for the reliability assessment, which improves accuracy through the second-order approximation and reduces computational cost. Jung et al. [ 21 ] proposed a method to estimate the optimal number of samples by considering the costs of optimizing the design variables and adding new samples. Hao et al. [ 20 ] achieved significant computational cost reduction by decoupling the reliability optimization and the confidence analysis. Furthermore, Jung et al. [ 23 ] proposed a sampling-based RBDO method that evaluates the confidence by estimating the input distributions using the multivariate kernel density estimation on bootstrap samples of the input data and calculating the reliability under each of the estimated distributions. In contrast, in the field of mathematical optimization, robust optimization (RO) [ 4 ] is used as an optimization framework that handles uncertainties. In RO, an uncertainty set is constructed as a set of the possible values of uncertain parameters, and the uncertainty is considered by solving a two-stage optimization problem that seeks the optimal decision variables for the worst-case uncertain parameter values within this uncertainty set. As an application of RO in the field of structural reliability, Ben-Tal and Nemirovski [ 6 ] optimized the design of antennas. Moreover, Kanno and Takewaki [ 24 ] proposed a method for the optimal design of trusses with uncertain external forces under a volume constraint, describing it as a nonlinear semidefinite programming problem and approximating it with sequential semidefinite programming. Furthermore, Kanno [ 26 , 27 ] proposed a robust optimization problem that conservatively approximates CBDO by constructing an uncertainty set for external forces using the order statistics, and minimized the volume of structures under compliance constraints. Distributionally robust optimization (DRO) [ 15 , 18 ] is one of the models that handle uncertainties and have been actively studied in recent years. While RO constructs an uncertainty set for the possible values of uncertain parameters, DRO treats uncertain parameters as random variables and constructs an uncertainty set, called the distributionally uncertainty set , for the probability distributions that they follow. By incorporating probabilistic aspects, DRO overcomes the excessive conservativeness of optimal solutions that is a challenge in RO. Several methods have been proposed for constructing the distributionally uncertainty set. Delage and Ye [ 15 ] proposed a moment-based method. They construct an ellipsoid using the sample mean and sample variance-covariance matrix calculated

Read the complete text