I’m Meenu, a Postdoctoral Research Fellow at Johns Hopkins University working at the intersection of computational mechanics and scientific machine learning — building solvers, surrogates, and optimization tools that help us understand how structures and materials really behave.

Physics, analysis, and design — in one workflow.

From phase-field fracture and large-scale topology optimization during my Ph.D. at IIT Roorkee, to evolutionary deep neural networks and neural-operator benchmarks at Johns Hopkins, my work is about making hard mechanics problems solvable at scale.

Computational Mechanics Scientific ML Phase Field Fracture Topology Optimization HPC & FEniCS
U. Meenu Krishnan

Recent work · 2026 preprints

Two new open-source projects

From my postdoctoral work at Johns Hopkins: a GPU-native, differentiable finite element framework, and a controlled benchmark for machine-learning surrogates of fracture.

Koyna Dam crack pattern — 96 million DOF, solved in 6.5 h on one GPU.

Preprint arXiv:2611.23682 · November 2026

Axion: A differentiable, GPU-native finite element framework, demonstrated on phase-field fracture

Wei Wang, Meenu Krishnan U., Somdatta Goswami

Axion (Autodifferentiable, XLA-compiled, Implicit, On-device Numerics) is a finite element framework written in JAX. The element kernels, Newton and Krylov iterations, bound-constrained updates, staggered coupling of fields and the loop over load increments are expressed as a single program that is just-in-time compiled and runs on the GPU without host intervention.

Consistent tangents come from forward-mode automatic differentiation of the element residual, and derivatives of the solution with respect to model parameters come from differentiating the same program. Operators are applied element by element without assembling a sparse matrix, octree refinement lives in fixed-size arrays, and a cell-based domain decomposition runs the same program on several GPUs. Phase-field fracture is the first module.

  • 37.9×faster than FEniCSx, one GPU vs one CPU (2D tension benchmark)
  • 92% / 96%strong / weak scaling on four GPUs
  • 96M DOFconcrete gravity dam on a single GPU

Preprint arXiv:2611.03473 · November 2026

Fracture-Bench: A Controlled Benchmark Suite for Machine Learning Surrogates in Phase-Field Fracture Mechanics

Meenu Krishnan U., Maryam Hakimzadeh, Indrashish Saha, Meer Mehran Rashid, Lori Graham-Brady, Somdatta Goswami

Which neural surrogates can you trust to replace phase-field fracture simulations, for which problems, and how far beyond their training data? Fracture-Bench is an open benchmark of two phase-field datasets with fixed splits, common test metrics and a description of each dataset’s physical regime. A Fourier neural operator (FNO), DeepONet, Latent DeepONet, a UNet and a conditional diffusion model are trained on identical splits and compared on accuracy, training time, inference time and model size.

Dataset 1 · 1,000 AT2 simulations

Multiple cracks growing from a notched hole in a hyperelastic plate under random non-uniform biaxial boundary displacements (44 load steps, 128×128 grid).

Dataset 2 · 4,000 PF-CZM simulations

A pre-cracked plate with two inclusions, each ten times stiffer or more compliant than the matrix (128×256 grid).

  • 10−6mean test error of FNO and Latent DeepONet on Dataset 1, about 10× below UNet and DeepONet
  • 706 s vs 6 msper prediction on Dataset 2: the diffusion model is most accurate (about one sixth of the UNet error), but far slower than UNet
  • Final stepsmost of the error appears after crack localization, so rankings depend on the task
Dataset 1, held-out sample, 44 load steps: finite element result next to two surrogate predictions.

Data, codes, splits, regime descriptors and generalization splits are public, with a protocol for adding under-represented regimes such as mixed-mode, three-dimensional and experimentally paired fracture.

Contact

I’m always happy to discuss research ideas, collaborations, or opportunities. Reach me at ukrishn4@jh.edu or fill in the form below.

Milestones

Recent Achievements

Johns Hopkins University campus

Postdoctoral Research Fellow, Johns Hopkins

2024 — Present

Developing evolutionary deep neural networks for multi-physics problems in solid mechanics, bridging mechanics and machine learning.

Learn more
Dynamic crack branching benchmark

Fracture-Bench: a benchmark for neural operators

Johns Hopkins · Ongoing

Shared datasets and consistent evaluation so neural-operator surrogates for fracture can finally be compared on equal footing.

Learn more
Large scale topology optimization result

Large-scale topology optimization, then 3D printed

Ph.D. · IIT Roorkee

Scaled optimization to large 3D problems in FEniCS with MPI parallelism, then printed the designs to check they really carry their load.

Learn more
Phase field fracture simulation

Adaptive phase-field fracture with automatic time stepping

Ph.D. · IIT Roorkee

Energy-based error indicators drive mesh refinement, keeping fast crack growth accurate without meshing the whole domain finely.

Learn more
Auxetic metamaterial deformation

Auxetic metamaterials by design

Ph.D. · IIT Roorkee

Designing material layouts with negative Poisson’s ratio through simulation, optimization, and fabrication.

Learn more
IIT Roorkee campus

Ph.D. in Computational Mechanics, IIT Roorkee

Doctoral research

Efficient algorithms for phase-field fracture and 3D-printable topology optimization: mesh adaptivity, time stepping, and parallel HPC.

Learn more

Selected work

Research Highlights