Open tools for research: data, simulation and monitoring
📍 Where we are: Part V · Beyond the Desktop — Chapter 15 of 16. Chapter 11 showed open machines built for research. Here we look at the open data, simulation codes and monitoring that make that research checkable.
Imagine a cooking contest where every cook publishes the recipe, the oven temperature log and a photo of the cake, so anyone can bake it again or build a model that predicts the outcome. Open research tools in additive manufacturing work like that: public measurements, free simulation software, and shared sensor recordings. The difficulty is the same as in cooking: a recipe is not the same as a result you can reproduce in your own kitchen.
What this chapter covers
We tour three kinds of open tools: data (NIST AM Bench and other public datasets), simulation (free codes for heat flow, fluid flow and stress), and monitoring (open software that watches a print). We end with reproducibility: what you must record so another person can repeat your work. Each named tool or dataset has a public source in the references.
Open data: benchmarks and datasets
NIST AM Bench is a series of benchmark measurements and challenge problems, run by NIST on a roughly three-year cycle (2018, 2022, 2025, with 2028 planned). Its stated aim is to give modellers rigorous, highly controlled measurement data so they can check their simulations, and all the data are archived for public use [1]. Modellers get the experiment description, predict the outcome blind, and submit results; the measurements are then published. The 2018 round is documented in a peer-reviewed outcomes paper, which describes AM Bench as established to provide rigorous measurement data for validating AM simulations across processes and materials [2].
The 2025 round has nine benchmark sets: eight on metals and one on polymers. Examples are nickel superalloy 625 built by laser powder bed fusion with different feedstock chemistries, fatigue of Ti-6Al-4V, laser hot-wire directed energy deposition with residual-stress measurements, and a photopolymer resin problem on cure depth versus light exposure [3]. Data hubs, a searchable catalog and free compute on JHU SciServer are linked from NIST [4]. The page I read does not state a licence or file format for the data, so check each data publication before reuse. NIST says the future focus will move to realistic components designed with industry, with first sets planned for 2028 [1].
Other public datasets I could verify:
- NIST registered in-situ and ex-situ dataset for metal powder bed fusion (NIST AMS 100-69, by Yang, Van Handel, Lu, Yeung and Lane, May 2025), which links process-monitoring data to X-ray computed tomography and other measurements [5].
- ORNL in-situ dataset from a laser powder bed build of 64 stainless steel 316H cylinders: layerwise visible and thermal images, scan paths, XCT, and fatigue results for 60 of them [6]. The page I read lists no licence.
- A melt-pool compilation ("Melt-Pool-Kinetics") in Scientific Data (2025) that gathers images from 32 datasets across 23 sources, 1.9 TB raw and released as a 48.6 GB HDF5 collection, according to the paper's abstract [7]. The paper itself notes that 75% of the datasets are DED (laser metal deposition and WAAM) and that only 7 distinct materials appear, so diversity is limited.
"Open" here means public to download, which is not always openly licensed. For machine-learning work the compilation is useful, but each source dataset keeps its own terms.
Open simulation codes
Simulating AM means solving heat flow (and often fluid flow and stress) while a heat source moves. Several free codes do this:
| Code | What it does | Basis and licence | Evidence |
|---|---|---|---|
| AdditiveFOAM | Heat and mass transfer, melting and solidification for metal AM; moving beams; can export thermal data to the ExaCA grain code | Built on OpenFOAM [8]; GPL v3 or later | JOSS paper, 2025 [9]; repo [10] |
| FEniCS | General finite-element toolkit; users write the equations | Open source library | Alnæs et al. [11]; used in a phase-field thermomechanical AM model (a short 2023 proceedings paper, so an example of use, not strong validation) [12] |
| MOOSE | Multiphysics finite-element framework | Open source framework | Framework paper [13]; used in a 2026 open-source WAAM thermo-mechanical study [14] |
A recent example of what an open stack can do: a 2026 Additive Manufacturing paper combined contour-method, neutron-diffraction and hardness measurements with MOOSE thermo-mechanical simulations of 316 L WAAM walls, and reports predicted stresses within 20-50 MPa of neutron data for a 3-layer wall [14]. That is one study on one geometry and material, not yet cited by others, so treat it as lab-scale evidence rather than a validated general tool.
Maturity differs. AdditiveFOAM is a purpose-built AM code with documentation and a peer-reviewed software paper; the repository notes that its 1.0.0 release supports OpenFOAM-10 while the current documentation targets a newer OpenFOAM, so check version pairing before you install. FEniCS and MOOSE are general tools: you must write or adopt a model, which is flexible but more work, and a published model is only as good as its validation.
A sanity check before heavy simulation
A classic analytical model by Rosenthal (1946) gives the temperature around a moving point heat source [15]. It is crude: the point source makes temperature blow up near the laser, and it ignores melting, but it shows the shape of the answer and catches unit mistakes. The inputs below are illustrative, not measured, and I ran the snippet with Python 3 and the standard library only.
import math
eta, P = 0.4, 200.0 # absorptivity, laser power [W]
k, alpha = 20.0, 5e-6 # conductivity [W/m/K], diffusivity [m^2/s]
v, T0 = 1.0, 300.0 # scan speed [m/s], preheat [K]
def T(xi, y=0.0, z=0.0): # xi = x - v*t (moving frame) [m]
R = math.sqrt(xi*xi + y*y + z*z)
return T0 + eta*P/(2*math.pi*k*R) * math.exp(-v*(R + xi)/(2*alpha))
for d_um in (500, 1000, 2000):
d = d_um*1e-6
print(f"{d_um:4d} um ahead: {T(+d):8.1f} K behind: {T(-d):8.1f} K")
Output: ahead of the laser the plate is still at 300 K, and behind it the temperature falls from about 1573 K at 0.5 mm to about 618 K at 2 mm. That comet tail is the physics a full code such as AdditiveFOAM resolves far better.
Open research loop: public data, open codes and open monitoring feed each other, and a reproducibility record closes the loop.
Original diagram by the authors, created with AI assistance.
Open monitoring
In-situ monitoring means watching the process as it runs. At the desktop end, Obico is an open-source platform (self-hostable) that uses a trained image model on webcam frames to detect failed prints and can pause them through OctoPrint, Moonraker (Klipper) or other connections [16]. At the research end, the open datasets above are the substitute for owning an instrumented industrial machine, and the reAM250 platform in Chapter 14 is designed for adding your own sensors. I did not find a single open standard that unites these two ends.
Reproducibility: what to record
A result is reproducible when someone else can repeat it. Minimum record: software name and version (and the OpenFOAM or library version it depends on), input files, material lot and settings, machine and firmware version, and the exact data release you used. AM Bench stresses controlled data and blind predictions for that reason [1]. Be wary of claiming "validated" from one dataset: the 2022 microstructure data had many downloads but few challenge submissions, which shows availability is not the same as use [1].
Why it matters
Closed industrial machines hide their data. Open data and codes let students and small labs test ideas without a million-dollar machine, and let reviewers check claims.
In the real world
Expect to see these in university and national-lab work: AM Bench for model validation, AdditiveFOAM or MOOSE for process simulation, and public monitoring datasets for machine-learning studies.
Key terms
- AM Bench — NIST's series of benchmark measurements and challenge problems for validating AM simulations.
- Challenge problem — a blind prediction task: modellers predict, then compare with measured results.
- OpenFOAM — an open-source toolbox for computational fluid dynamics.
- AdditiveFOAM — an ORNL code, built on OpenFOAM, for heat and fluid flow in metal AM.
- FEniCS — an open-source toolkit for solving partial differential equations by finite elements.
- MOOSE — an open-source multiphysics finite-element framework.
- In-situ monitoring — sensing a print while it is being made.
- XCT — X-ray computed tomography, used to image internal pores after a build.
- Reproducibility — the ability of others to repeat a result from the published record.
Where this leads
Last, we turn from tools to people: Contributing, community and the honest verdict.