Zahra Pak

Topic

Toward Interoperability Between FEM Systems: A Domain-Aware Framework for Translating and Benchmarking FEBio Models in PolyFEM

Department of Computer Science

Date & location

• Thursday, May 14, 2026

• 2:00 P.M.

• Engineering Computer Science Building

• Room 468 and Virtual Defence

Reviewers

Supervisory Committee

• Dr. Teseo Schneider, Department of Computer Science, University of Victoria (Supervisor)

• Dr. Jaya Prakash Champati, Department of Computer Science, UVic (Member) 

External Examiner

• Dr. Kin Fun Li, Department of Electrical and Computer Engineering, University of Victoria 

Chair of Oral Examination

• Dr. Mario Bras, Department of Mechanical Engineering, UVic 

Abstract

Finite element method (FEM) software is central to computational engineering, scientific computing, and biomechanics, yet the practical reuse of simulation models across FEM platforms remains limited by differences in data representation, modeling assumptions, and solver-specific semantics. This thesis investigates that interoperability problem through a concrete and practically relevant case study: translation from FEBio, an XML-based plat form designed for nonlinear and multiphysics biomechanics, to PolyFEM, a modern FEM framework that exposes a JSON-driven simulation interface.

The thesis presents a domain-aware software framework for extracting, organizing, partially translating, and benchmarking FEBio models in a PolyFEM-oriented representation. The proposed system is implemented as a modular Python package consisting of a structure-aware FEBio parser, an intermediate representation, a translation layer that produces PolyFEM-compatible output, a writer that generates mesh and metadata artifacts, and a benchmark runner that evaluates both conversion coverage and PolyFEM execution success over a dataset of 126 FEBio models.

The system classifies each case into one of four categories: success (all materials map directly to PolyFEM equivalents and PolyFEM executes the result correctly), partial (the model converts but one or more materials have no direct PolyFEM equivalent and are mapped to supported fallbacks with discrepancies documented), unsupported (the FEBio analysis module has no corresponding PolyFEM physics formulation), and failed (a code-level parsing error prevented conversion). Of 126 benchmark cases, 26 convert fully successfully, 20 convert partially with documented material fallbacks, 78 are correctly classified as unsupported due to fundamental physics gaps between the two platforms, and 2 fail due to parsing errors. Of the 46 cases where a sim.json was generated, 39 (85%) execute successfully in PolyFEM.

The results reveal that the primary barrier to interoperability is not syntax translation but physics scope: PolyFEM does not implement fluid-structure interaction, Biot poroelas ticity, or reaction-diffusion formulations, which together account for 78 of the 126 benchmark cases. Within the subset of solid-mechanics models that fall within PolyFEM’s supported physics scope, the conversion pipeline achieves an 85% PolyFEM execution success rate, demonstrating practical effectiveness for the supported domain. The study also introduces the concept of layered interoperability — a structured distinction between structural transfer ability (geometry, connectivity, material metadata, surface information, control parameters) and semantic transferability (constitutive laws, multiphysics coupling, contact behavior, rigid body mechanics) — as the conceptual framework that makes the benchmark results inter pretable as research findings rather than mere software metrics.

This thesis makes four main contributions. First, it provides a working, modular software artifact for FEBio parsing, PolyFEM-compatible output generation, and automated PolyFEM execution validation. Second, it introduces a four-status benchmark methodology for evaluating conversion coverage and execution success across heterogeneous FEBio examples. Third, it offers a detailed analysis of where structural translation succeeds, where physics scope limits applicability, and where semantic mismatches produce documented partial results. Fourth, it establishes a foundation for future work on solver-aware intermediate representations, richer material translation, and broader physics coverage across FEM plat forms.