In this post we demonstrate the features of high resolution topology optimization through the demonstration problem offered by GE in their jet engine bracket challenge.
Computing power in the cloud now enables topology optimization with a resolution matching that of modern metal additive manufacturing machines (i.e. 0.1 mm for parts around 10 cm). The advantage is twofold: you fully exploit the geometric freedom of AM, and you avoid time consuming manual post-processing of the design, such as smoothing or complete CAD redrawing.
As an example, let us look at the jet engine bracket challenge, which GE announced in 2013. The challenge is simple: with five given mounting interfaces, design the best bracket that can withstand 4 different load conditions.
For this example we defined a target weight for the bracket of 300 g, which is approximately 10 % lighter than the challenge winner design. Our highly parallelised algorithm returned a printable final design, which stayed comfortably below the maximum stress limits. To give an impression of the resolution, the figure below shows a zoom of a cut plane, which reveals the individual elements making up the small rod on the downside of the part. This resolution is already smooth in a manufacturing sense. Further smoothing is typically not required (but very easy and geometrically robust to apply).
The amount of elements in the rod cross-section could easily have resolved a hollow rod, if that had been the optimal result. Another noteworthy feature is the prevalent use of thin walls. Most commercial topology optimisation software simply can't resolve thin walls, and their algorithms converge towards thick rod structures. This is a result of the coarse mesh, and in general one have to consider the mesh resolution together with the manufacturing resolution to make the optimal design.
Looking through the entries of the challenge, one quickly realises that only a few of the entries are applying topology optimisation. Even the winner is a manual design effort. This is curious, since the challenge is clearly formulated as a topology optimisation problem. The suggested designs submitted for the competition emphasize the issues of current topology optimization methods: they are simply too difficult to use and the necessary final manual redesign effectively renders the method almost irrelevant. High resolution topology optimization is a great way to resolve this issue, since it naturally leads to manufacturing-ready designs.
Update: See the follow up post, which answers some popular questions.
Written by Klaus Loft Højbjerre ·