Suspension bridges are reaching their limits. Despite the enormity of many existing suspension bridges like the Golden Gate Bridge, ambitious plans for far longer suspension bridges in countries like Italy and Norway are approaching or exceeding the physical constraints of suspension bridges. At the same time, the enormous carbon footprints of these bridges (largely stemming from the production and transport of materials) are beginning to be taken as serious planning considerations. Now, a team of researchers from the Technology University of Denmark (DTU) are reimagining the suspension bridge – and they’re using supercomputing to do it. 

“Above a main span of three kilometers, the bridge girder’s self-weight is quickly becoming the governing load, which means that the bridge cannot carry much else than its own weight,” said Mads Baandrup, a former PhD student at DTU who now works for engineering company COWI, in an interview with the Partnership for Advanced Computing in Europe (PRACE). “To build longer bridges, we have to rethink the design entirely.” 

So Baandrup worked with DTU professors Ole Sigmund and Niels Aage to do just that, reinventing the design of the traffic-bearing component of suspension bridges (the “bridge deck”). They achieved this using a computational technique called topology optimization, which, due to its high computational footprint, has more commonly been used for smaller-scale applications like aircraft wing design. “With the recently increased power of supercomputers,” Baandrup said, “we could adjust the method to apply it to large-scale structures.”

The research team analyzed a bridge segment five meters high with a surface area of 2,250 square meters. Since this section could effectively be copied and pasted for the remainder of the bridge, this segment was used as a representative sample. The segment was divided into 2 billion voxels (each representing just a few centimeters of real-life space). 

Then, the researchers “voided” the existing bridge structure, erasing anything that could point to conventional bridge designs. Starting from that blank slate, the topology optimization tool decided whether each voxel should be air or steel for maximum structural integrity. “In this way, the optimized structure is calculated from scratch,” Baandrup said, “without any assumptions about what it should look like.”

The optimization was run on the Joliot-Curie supercomputer, a PRACE supercomputer hosted by GENCI in France. Joliot-Curie is divided into four partitions: an Intel Skylake partition; an Intel Knights Landing partition; an AMD Epyc Rome partition; and an Intel Cascade Lake partition. When last benchmarked, Joliot-Curie delivered 1.3 Linpack petaflops, placing it 417th on the most recent Top500 list of the world’s most powerful publicly ranked supercomputers. The full calculation took 85 hours across 16,000 nodes.

The result? An internal bridge girder design that looks rather alien, with rippling waves of reinforcement running through it rather than the minimal geometric approach used since the 1950s. While this structure is both eye-catching and ostensibly stronger than existing approaches, its feasibility is in question. “The software identifies the optimal structure but does not take into account if the structure is actually buildable,” Baandrup said. So the researchers modified the design into something more practically buildable using current manufacturing methods, using a range of thinner, curved steel plates. There, the researchers found another advantage: 28% lower steel consumption. 

“Our results reveal a huge potential in rendering construction more ecological,” Baandrup said. “In the future, the construction industry should not only think about how to reduce cost but also how to reduce energy consumption and CO2 emissions. With our results, we believe we can initiate this discussion.”

To read PRACE’s reporting on this research, click here.

Header image: The new bridge concept applied to the 2682-meter Osman Gazi bridge in Turkey. Image courtesy of the researchers.