Shaking up seismic analysis with 2D models
While earthquakes are a big problem, seismic engineers still use an analysis model from the 1970s. The field needs a modern model that can accurately represent how waves travel through soil. To model the randomness of soil properties, a large number of simulations, each taking an hour or more to run, must be conducted. Hence, there is a need for optimising both OpenSees execution and the workflow controlling the thousands of simulations on NeSI platforms.
University of Canterbury researchers Prof. Chris McGann and PhD student Chris de la Torre collaborated with NeSI research software engineers Alex Pletzer, Chris Scott and Mandes Schönherr to optimise 2D and 3D models they developed that will enable seismic engineers to simulate micro ripples of seismic waves through soil.
A model that runs 1.7 times faster and which will save the researchers 100,000s of core hours. The new model will help engineers better understand how soil will affect building foundations during earthquakes. This will lead to more resilient buildings, reducing earthquake property damage and potentially saving lives.
Earthquakes are an everyday occurrence in New Zealand. Around 15,000 earthquakes occur around the country each year. Most are too small to be noticed, but larger ones can be deadly, as the 2011 Christchurch earthquake showed. It’s therefore vital that New Zealand understands its earthquakes, to build resilient structures.
Professor Chris McGann and PhD student, Chris de la Torre are researchers at University of Canterbury and members of QuakeCoRE, the New Zealand earthquake resilience centre. Together, they are making a better seismic site response analysis model.
This model shows how an earthquake will affect an area’s soil. Different types of soil can greatly affect the amount of shaking during an earthquake. For instance, a building atop water-logged soil can find itself swallowed in quicksand during, while sandy soils can damage a building’s foundation.
The 1-dimensional model used by seismic engineers today has not changed much since its invention in 1970. This model simplifies soil layers so that each layer is a line of one soil type that the seismic wave travels through.
“We predict how particular soil deposits will modify shaking during an earthquake. This is usually all done in 1D analysis, with every soil layer modelled as homogenous and stretching to infinity,” Chris de la Torre said.
In reality, soil layers can have multiple soil types that can scatter seismic waves and cause micro ripples. The pair set out to design a new multi-dimensional model to better capture this.
“The 1D analysis everyone uses is very cheap and portable – you can use it on a laptop. But it was developed with the limitations of 1970s computing in mind, which we no longer have. We’re using 2D and 3D to randomise the soil layers so we can observe wave scattering. That happens in real life and you can’t capture that in 1D,” Chris de la Torre said.
But there are challenges with this upgrade.
Because of the computational requirements, the pair rely on NeSI's HPC platform and the Māui supercomputer to run their test models. Their plan was to create a 3D model, but after testing they found running a 50x50 3D model took over 150 core hours. Adding more cores did little to reduce the model run-time, as it reached peak scalability at 20-40 cores.
“The consultancy helped us realise the limitations of the tool we were using. When we expanded this to 2D and 3D domains, we reached the point where we couldn’t run on local computers anymore. We had to use NeSI supercomputers to reduce the analysis time,” Chris McGann said.
With advice from NeSI research software engineer, Alexander Pletzer, the pair decided to focus on the 2D model. This would still be a sizeable step up from the 1D model, but simulations could still run quickly on Māui.
“The ones I do now take less than 30 minutes using eight nodes. This is great because we need 30 realisations for a parameter study,” Chris de la Torre said.
“With each parameter study, we then see what happens when we introduce soil randomisation in ground shaking. This means we’re calculating 240 permutations, times 30 realisations just to wrap our heads around it. We’re streamlining this but it still takes a lot of computing power,” Chris McGann said.
The Consultancy project with NeSI helped the researchers streamline their code to reduce run-times. They also received advice on how best to arrange jobs for the multiple cores. This led to a conference paper published by the pair and co-authored by Alexander.
“The consultation helped us make sure the computations were running in parallel, so we got the most out of the high-performance computing resources. We’re not computer scientists, we know just enough to be dangerous, so it was really useful to have people who were experts on computer architecture to help us compile the programs,” Chris McGann said.
The next stage of the pair’s project is adapting the model they’ve built on Māui to allow earthquake engineers to use it in the field. This means finding a way to represent soil randomisation so engineers can run on smaller computing set-ups.
“There’s a potential to use this to improve 1D analysis, but ideally we want people to take that leap beyond 1D. We’re looking at manual partitioning and changing to another OpenSees software to get beyond our current software limitations,” Chris de la Torre said.
“NeSI was great. The team was really excited to help us out. It felt like they were part of our research team, not just people we were asking to work for us. They got their heads into it,” Chris McGann said.
The pair’s work will help new buildings survive earthquake damage and improve decades-old soil analysis methods. With NeSI’s help, the pair are building a foundation for an earthquake resilient New Zealand.
Do you have an example of how NeSI support or platforms have supported your work? We’re always looking for projects to feature as a case study. Get in touch by emailing email@example.com.