From Biomechanical FEA to Geotechnical Numerical Modelling
I spent several years building finite element models of porous biological tissue before I built one of a slope. The first time a senior geotechnical colleague looked over my shoulder at a PLAXIS model and asked how I’d chosen the permeability ratio between layers, my instinct was the same one I’d used on soft tissue models years earlier: don’t trust a single value, run the model across a plausible range, and see whether the conclusion changes. It usually does, and that’s the point.
Where the two fields actually diverge
Biological tissue and soil are both porous media, but the similarity has limits I try not to overstate. Tissue behaviour is often close to isotropic over the length scales I modelled, with material variability coming mainly from biological differences between samples. Soil is the opposite: variability comes from geological structure, layering, and history — anisotropy, fissuring and stress history matter far more than they did in my earlier work. Carrying over a biomedical mindset without adjusting for this is a real risk, and I’ve had to unlearn some instincts as much as apply them.
What carries across
What does transfer cleanly is the discipline around the model, not the material model itself: formulating the boundary conditions before touching the software, deciding in advance what "success" for the model looks like, and treating every parameter as a range rather than a point value until sensitivity analysis says otherwise. Model validation habits also carry over directly — in research, a model is not accepted until it reproduces independent experimental data; in consulting, I hold geotechnical models to the same standard against site monitoring or back-analysis of known behaviour wherever that data exists.
A concrete example
On a settlement assessment for underground works, I built the ground movement model the way I would have built a tissue deformation model: define the governing mechanism first, then decide which parameters the outcome is actually sensitive to, rather than refining every input equally. That approach flagged early that the predicted settlement was far more sensitive to the assumed stiffness profile with depth than to the exact permeability value the team had spent the most time debating — which redirected our effort to where it mattered.
Where I’m careful
I’m wary of engineers — including earlier versions of myself — who assume computational fluency in one domain transfers wholesale to another. It doesn’t. What transfers is the research habit of questioning the model before trusting the output. The material science, the failure modes and the consequence of getting it wrong are genuinely different, and I treat that difference as the starting point, not an afterthought.