Computational engineers seeking to design lightweight, high-performance load-bearing components have long used structural optimization routines to carve away unnecessary mass while preserving overall structural stiffness. However, practical engineering components often demand distinct materials capable of handling opposing structural roles, specifically utilizing materials that excel in tension alongside complementary materials tailored for compressive resistance. To address this need in an open and computationally reproducible framework, researchers Hongin Kim, Jaehee Choi, Minsoo Sung, and Donghyuk Jung have introduced MBESO2D, an open-source MATLAB implementation of a stress-state-based bi-directional evolutionary structural optimization formulation. Documented in the journal SoftwareX, the software bridges a longstanding reproducibility gap by transforming established discrete multi-material optimization equations into an inspectable, fully documented codebase that operates entirely within standard numerical computing environments without requiring specialized external toolboxes.
The underlying mechanics of the software trace back to the multi-material bi-directional evolutionary structural optimization formulation originally outlined by Li and Xie. In classical continuum topology optimization, methods such as the Solid Isotropic Material with Penalization technique apply continuous material density variables across a discretized finite-element grid. By contrast, bi-directional evolutionary structural optimization operates via discrete updates, systematically removing inefficient solid elements and restoring void elements over iterative analysis steps. While earlier educational software releases successfully popularized single-material evolutionary approaches, open implementations for multi-material discrete systems remained scarce. The theoretical framework developed by Li and Xie provided a clear mathematical strategy wherein overall material utilization dictates whether an element remains solid or becomes void, while an explicit local stress-state indicator determines whether retained elements are designated as tensile or compressive materials. Because source code mapping those equations directly to execution routines was previously unpublished, independent verification and direct methodological comparisons remained challenging for the computational mechanics community.
MBESO2D realizes this multi-material optimization workflow across two-dimensional design spaces discretized into regular four-node bilinear quadrilateral plane-stress finite elements. Across the computational mesh, each element receives both an internal density variable that switches between solid status and a minimal residual ersatz value, as well as an explicit material identifier corresponding to void space, tensile material, or compressive material. Within the finite-element analysis routines, void regions are not physically excised from the numerical equations; instead, following an established soft-kill convention, they maintain a very small residual stiffness proportional to a penalization exponent. This residual stiffness ensures continuous numerical stability across the global degrees of freedom, preserves weak structural pathways, and provides a continuous finite-element response field that enables previously eliminated void zones to be evaluated and potentially restored during subsequent evolutionary iterations.
The mathematical engine of the software calculates the mechanical response at every design iteration by solving the global linear-elastic equilibrium system under prescribed boundary conditions and applied mechanical loads. From the resulting displacement solutions, the software evaluates the total compliance or strain energy, which serves as a global response tracker and convergence metric rather than a directly minimized objective function. To assign specific material roles across the domain, the program computes a local stress-state indicator defined as the sum of the two in-plane principal stresses at each element. A positive sum indicates a tension-dominant local stress environment, prompting the assignment of the designated tensile material, whereas a negative sum identifies a compression-dominant state that triggers the assignment of the compressive material. Alongside this stress indicator, an internal material utilization score is calculated from the centroidal von Mises stress normalized by the allowable strength limits of the respective materials.
A critical algorithmic mechanism built into the software governs how material utilization scores transition across iterations through a dynamically updated normalization parameter termed nk. When solid volume fractions are progressively reduced according to a fixed evolutionary ratio, structural elements composed of high-strength tensile material could naturally exhibit lower raw stress ratios than adjacent compressive elements, risking premature and unintended elimination in early cycles. To stabilize this evolutionary trajectory, the software gradually shifts the tensile-side normalization baseline from the compressive allowable limit toward the actual tensile allowable limit across the projected volume-reduction steps. Before solid retention choices are finalized, both the local stress-state indicators and the utilization scores undergo spatial distance-weighted filtering across independent filter radii. Furthermore, the filtered utilization scores are combined with values from the preceding step in an explicit history-averaging calculation, dampening spatial oscillations and ensuring continuous, mechanically rational load paths.
Architecturally, MBESO2D is deployed across two primary standalone MATLAB solvers that share identical argument interfaces while addressing distinct engineering constraints. The baseline solver handles open, unrestricted design domains, whereas an extended solver incorporates designated boundary masks to protect fixed-solid regions and enforce non-design void exclusions. The extended solver also integrates automated self-weight assembly, dynamically updating gravitational body forces at each iteration as material types and solid-void boundaries shift across the grid. To maintain numerical efficiency and transparent source-level inspection, the finite-element matrix assembly utilizes vectorized indexing adapted from compact single-material evolutionary codes, allowing standard benchmarks to execute rapidly on standard personal computing hardware without relying on precompiled C binaries or proprietary solver extensions.
The research team evaluated the fidelity and numerical performance of the software across classic benchmark problems, beginning with a short-cantilever structure subjected to an asymmetric shear point load. In this benchmark, the software assigned an elastic modulus of 200 gigapascals and an allowable tensile stress of 250 megapascals to the tensile material, paired with a compressive material exhibiting a modulus of 20 gigapascals and an allowable compressive stress of 50 megapascals. While the resulting structural topology successfully developed characteristic tensile ties and compressive struts mirroring published patterns, detailed numerical comparisons revealed notable local variations. The optimized compliance reached 574.78 newton-millimeters, which was 6.41 percent lower than the reference value of 614.16 newton-millimeters. Simultaneously, the maximum centroidal von Mises stress reached 128.91 megapascals within MBESO2D, substantially exceeding the 53.32 megapascals documented in the original literature. This substantial local stress variation highlighted how subtle implementation differences in filtering and discrete element updates can influence localized peak stresses while maintaining comparable global structural compliance.
To rigorously investigate whether mesh density was driving these localized stress disparities, the authors conducted parametric mesh-refinement trials on the short-cantilever benchmark using grids of 50 by 30, 100 by 60, and 200 by 120 elements. As the mesh was refined, the full-domain peak von Mises stress escalated dramatically from 85.61 megapascals to 261.36 megapascals, directly demonstrating severe mesh sensitivity around singular boundary restraints and concentrated point loads. However, when the authors evaluated stresses outside localized 40-by-40-millimeter exclusion zones surrounding the fixed corner supports and the external load application point, the peak stresses stabilized into a tight, consistent band of 70.69, 74.36, and 72.65 megapascals across the three meshes. Although these exclusion-based stress evaluations mitigated mesh singularity sensitivity within the software, they remained consistently above the originally published reference values, demonstrating that differences in local stress sampling methodologies must be carefully documented when comparing evolutionary algorithms.
Further structural demonstrations focused on complex bridge-like domains featuring permanently retained solid deck regions subjected to a combination of distributed deck loading and active self-weight calculations. Case III(a) evaluated an arch configuration with a lower retained deck tie, whereas Case III(b) examined an inverted system featuring an upper deck and a lower suspended tensile belt. In both bridge configurations, the software preserved perfect vertical static equilibrium, yielding residual force imbalances relative to total downward gravity loads below 1.1 times 10 to the negative eleventh power. Reconstructed strain-energy verifications confirmed that compliance values derived from separate load decompositions matched full coupled solutions within 2.0 times 10 to the negative eleventh power. Although the computed compliance values were roughly one-third of the values reported in the historical literature—likely reflecting differing assumptions regarding out-of-plane thickness or self-weight distributions—the internal physical equilibrium and energy consistency of the solver routines remained mathematically unassailable.
To confirm that the internal finite-element calculations were strictly accurate, the authors performed independent cross-verification against the commercial finite-element package Abaqus/Standard 2024. Reanalyzing the exact final topologies of both the short-cantilever and the inverted bridge benchmark using four-node plane-stress elements under identical boundary constraints yielded absolute agreement: peak centroidal stresses, critical nodal displacements, and total compliance values differed by less than 0.01 percent between MBESO2D and Abaqus. Additional element-tracking studies confirmed that material-switching events, spatial filtering operations, and solid-retention decisions operated precisely as defined by the governing mathematical expressions. By providing an open, fully verified computational baseline released under an open-source license, MBESO2D supplies the academic community with a transparent foundation for educating engineering students, validating new multi-material stress criteria, and developing robust structural optimization workflows.
Subject of Research: Open-source computational implementation and numerical verification of stress-state-based multi-material topology optimization in MATLAB
Article Title: MBESO2D: an open MATLAB implementation of stress-state-based multi-material BESO
Article References: Kim, H., Choi, J., Sung, M., & Jung, D. (2026). MBESO2D: an open MATLAB implementation of stress-state-based multi-material BESO. SoftwareX, 36, Article 103231. https://doi.org/10.1016/j.softx.2026.103231
Image Credits: AI Generated
DOI: 10.1016/j.softx.2026.103231
Keywords: topology optimization, multi-material BESO, structural stiffness, stress state indicator, finite element analysis, compliance minimization, strain energy, MATLAB implementation, soft kill ersatz material, reproducible research, von Mises stress, spatial filtering, self weight optimization, plane stress elements, retained solid masks, computational mechanics
Cite Scienmag News
Reid Dalton. (November 20, 2026). Open MATLAB Code Automates Multi-Material Topology Optimization Based on Local Stress States. Scienmag. https://scienmag.com/open-matlab-code-automates-multi-material-topology-optimization-based-on-local-stress-states/
Reid Dalton. "Open MATLAB Code Automates Multi-Material Topology Optimization Based on Local Stress States." Scienmag, 20 November 2026, https://scienmag.com/open-matlab-code-automates-multi-material-topology-optimization-based-on-local-stress-states/. Accessed 20 November 2026.
Reid Dalton. "Open MATLAB Code Automates Multi-Material Topology Optimization Based on Local Stress States." Scienmag. November 20, 2026. https://scienmag.com/open-matlab-code-automates-multi-material-topology-optimization-based-on-local-stress-states/








