A structural design program can put eight beams through one joint if that arrangement meets its mathematical objective. The engineer who has to detail that connection may only spot the problem when the run is over. Researchers at MIT have tested a way to bring that engineer into the calculation early enough to influence where the material goes.
Their program looks at an evolving design as an image and highlights regions someone might want to revise. The team trained it with synthetic examples of two preferences. One sought the longest member in a structure. The other sought the most complicated connection. An engineer can use a suggestion to direct the next design change while the optimizer continues to work on the rest.
In one example the intervention improved the calculated linear buckling load by 39 percent compared with a conventional optimization run, at a cost of 15 additional seconds of design time. That figure concerns the onset of instability in a modeled structure. The researchers separately demonstrated changes to make connections easier to manufacture.
Where a drawing meets the shop floor
A drawing can satisfy a load requirement and still give a fabricator too little room to join the pieces. Asking the engineer about one troublesome region makes that disagreement visible while alternative arrangements are still available. The training examples capture only two recognizable preferences. A shop whose decisions turn on tool clearance or the order of assembly would have different reasons to flag a joint.