Prototype
AI Engineering & Simulation
AI-Assisted Engineering
An AI layer around established simulation tools such as ANSYS and CalculiX. It does not replace solvers or engineering judgement.
All projects
3D-Printer Vibration Analysis
AI-Assisted Engineering case study
Local AI reasoning plus deterministic engineering computation: a 62-part desktop 3D printer analysed for its first 10 natural vibration modes, with validation status reported in full.
What it is
A proof of concept combining local AI reasoning with deterministic engineering computation, to analyse the natural vibration of a desktop 3D printer.
It is presented as an R&D-stage engineering workflow, not a validated predictive product.
Problem addressed
Simulation workflows involve preparing models, checking inputs, running solvers and explaining results. This proof of concept tested whether a local AI model could assist those steps while an established solver performed the numerical calculation.
What has been built or demonstrated
- Reconstructed a 62-part desktop 3D printer from public drawings.
- Used a local Qwen 3.8 27B model to check input data and support engineering reasoning.
- Used CalculiX to calculate the first 10 natural vibration modes across mesh sizes and support stiffnesses.
- Used Qwen to generate the engineering explanation and HTML report.
- Identified rubber-foot stiffness as the dominant uncertainty.
Workflow
- 01Engineering Model
- 02Local AI Reasoning
- 03CalculiX Solver
- 04Numerical Results
- 05Interpretation & Uncertainty
Reported evidence
62
parts modelled
10
natural modes
24/24
automated tests passed
Software checks of the workflow, not physical validation.
11/11
AI-reported numerical values matched the solver output
Validation status
Status · Experimental validation pending
The three lowest vibration shapes matched the published study, but the calculated frequencies were 18–49% lower than the published measurements.
Rubber-foot stiffness was identified as the dominant modelling uncertainty. Its contribution to this difference has not yet been confirmed experimentally.
This is presented as an R&D-stage engineering workflow, not a validated predictive product. Experimental validation by Aryaveer is pending.
Why software checks are not validation. Passing software checks shows the workflow runs correctly; it does not show the model matches physical reality. That requires experimental validation.
Limitations
- Not physically validated: calculated frequencies were 18–49% lower than the published measurements.
- The model was reconstructed from public drawings.
- Rubber-foot (support) stiffness is the dominant modelling uncertainty.
- The 24/24 automated tests check the software workflow, not physical accuracy.
Planned next steps
- Experimental validation by Aryaveer (pending).
- Reduce the support-stiffness uncertainty and compare the results with measurements again.
Related projects
Interested in a pilot or technical collaboration?
We are open to pilot partners. Tell us about the engineering problem you are working on.