Research on Computational Solid Mechanics Lab @ UCI

Data-Driven Optimization of Advanced Casting Processes

Mon, 01 Jan 0001 00:00:00 +0000

Abstract

The production of advanced cast components, such as those used in aerospace turbine applications, is limited by process variability that reduces yield and consistency. Many parameters can be monitored and adjusted along the casting chain, but not all of them are equally relevant to final part quality. This project, part of the Pratt & Whitney Center of Excellence for Solidification Science at UCI, identifies and prioritizes the parameters with the highest influence on part quality, develops data-driven models that quantify process sensitivities, and establishes a roadmap for optimizing them. Ensemble Bayesian networks trained on process data reveal which variables drive each type of defect, and Bayesian optimization then selects, within safe operating ranges, the settings that minimize the probability of a defect, so that a few informative production trials replace exhaustive testing. The methodology is being developed on wax injection, the first stage of the investment-casting process, and is designed to extend stage by stage to the entire casting workflow.

Energetic Mesh Smoothing

Mon, 01 Jan 0001 00:00:00 +0000

Abstract

This project treats mesh improvement as an energy minimization problem. Each element of a finite element mesh is paired with an ideal reference element, and a hyperelastic pseudo-strain energy penalizes both its distortion and its change of volume, so that general-purpose finite element solvers can be used to smooth the mesh. A family of pseudo-energy densities inspired by the Seth–Hill generalized strains can be tuned to penalize disproportionately the worst elements in a mesh, which are the ones most detrimental to simulation stability, and grows without bound as an element degenerates, so that element inversion is impossible by construction. The same energy drives discrete topological operations, such as element swaps and edge splits and collapses, and accommodates spatially varying size fields and anisotropic adaptation through a modified metric. On three-dimensional meshes containing highly distorted elements, the method substantially improves mesh quality and consistently outperforms established smoothing techniques from the Cubit meshing library.

Impact of Solid Propellant Microstructure on Effective Behavior: a Data-Driven Perspective

Mon, 01 Jan 0001 00:00:00 +0000

Abstract

This research seeks to elucidate the relationships between microstructural features at the mesoscale and the effective thermomechanical behavior of solid rocket propellants through a novel data-driven multiscale modeling framework. Conventional approaches to microstructural homogenization, including theoretical and phenomenological models, often depend on restrictive assumptions that limit their applicability to complex, heterogeneous materials, while direct numerical simulations, though accurate, are computationally intractable for large-scale applications. Even advanced surrogate models trained on simulation data face limitations in generalizability and interpretability, as identical microstructural parameters can yield markedly different effective responses due to uncharacterized microstructural variability. The central hypothesis of this work is that interpretable, low-dimensional features governing the effective behavior of heterogeneous materials can be discovered via machine learning models trained on high-fidelity simulation data. To this end, we integrate mesoscale finite element simulations of representative volume elements with symmetry-preserving neural network architectures conditioned on latent representations extracted from microstructural images. Interpretability is achieved through a combination of spatial attribution techniques and correlation analysis between latent variables and physical descriptors such as porosity, inclusion morphology, and distribution metrics. The resulting framework yields computationally efficient, physically informed surrogate constitutive models capable of predicting the behavior of solid propellants across a broad design space, generating new knowledge on the core features of a microstructure that dictate its effective behavior, while also enabling scalable, high-fidelity simulations at the engineering scale.

Predictive Discovery of Radiation-Resistant Tungsten Alloys for Extreme Fusion Environments

Mon, 01 Jan 0001 00:00:00 +0000

Abstract

Plasma-facing components in fusion reactors must survive intense neutron irradiation and extreme heat loads, and tungsten alloys are the leading candidates for the task. This multi-campus University of California project establishes a hub for the predictive discovery and accelerated demonstration of durable, supply-resilient alloys for fusion reactors, combining multiscale modeling, machine learning, and advanced irradiation and performance testing. The Computational Solid Mechanics Lab participates as a co-investigator, contributing to the mechanics modeling of these materials.

Quantum Computing for the Simulation of Heterogeneous Materials

Mon, 01 Jan 0001 00:00:00 +0000

Abstract

Predicting whether a structural part will bend, crack, or hold requires resolving the internal structure of the material it is made of, and the cost of such simulations grows steeply with resolution, until even the largest supercomputers fall short. Quantum computers offer a fundamentally different scaling, but so far only simplified versions of materials simulations have been written as quantum circuits. Through the UCI–LANL–SoCalHub Research Fellowship, which pairs UCI doctoral students with scientists at Los Alamos National Laboratory, this project translates the simulation of elastic deformation in heterogeneous materials, such as composites, into quantum algorithms. A general elastic formulation has been implemented as a quantum circuit and validated against a classical benchmark, a particle embedded in a surrounding matrix, with which it agrees. Co-mentored by Prof. Rimoli and Ricardo Lebensohn of Los Alamos, the work is a step toward simulations that classical machines cannot afford and, ultimately, toward lighter and more damage-tolerant aerospace structures.

Tensegrity Metamaterials for Extreme Energy Absorption

Mon, 01 Jan 0001 00:00:00 +0000

Abstract

Tensegrity structures are assemblies of slender members in which pre-stressed tensile cables form a continuous network while compression bars remain isolated, so that each member keeps its sign of loading whatever the applied load. Built as periodic lattices, they become metamaterials with a property that no other known material system exhibits: instead of failing by localization, as a fracture surface, a shear band, or a collapsing layer of cells, they spread deformation through the whole volume, with a recoverable, foam-like stress plateau, and absorb orders of magnitude more energy than lattices of the same density. This project seeks the fundamental mechanisms behind that delocalization, with particular focus on the extreme nonlinear regime of severe deformations and deformation rates typical of blast exposure, and aims to distill them into design rules for metamaterials for extreme energy absorption. The working hypothesis is that delocalization is topological: it stems from the connectivity of the lattice rather than from its geometry. Our first results support it. Representing each lattice as a pair of graphs, the tension and compression networks, deformation delocalizes whenever the tension network remains more connected than the compression network, and graph-theoretic connectivity measures capture the transition quantitatively.