Article by Jeffrey van der Gucht
At Windbase we are always trying to innovate our methods and services. To do so, we regularly collaborate with universities. One of our latest collaborations was with the Universiteit Twente (UT) master student in Construction Management & Engineering, Qinshuo Shen.
The goal of Shen’s master research was to develop a new and fast method to gain insights in the structural behavior of a wind turbine foundation. Within Windbase we use Finite Element Analysis (FEA) to obtain these insights. Although our FEA models are very detailed, they typically have very large calculation times. This is a bottleneck in our production process and limits the amount of iterations we can do to deliver the most optimal designs. Having access to fast estimates of such detailed calculations allows our designers to do some of the optimization beforehand, leading to better designs in less time.
To tackle this problem, Shen proposed to use machine learning to create a meta-model which can estimate key features in the outcome of a FEA calculation within fractions of a second. However, before he could create a machine learning algorithm, he first had to create a (somehow standardized) data set. The models of our foundations contain hundreds of parameters, which was out of scope for the graduation project of Shen. Therefore, he had to reduce the amount of free parameters in a clever way. With this reduced set of parameters, he began to create a large data set of FEA calculations. During this data generation phase, our FEA computers were running day and night, using all the downtime there was between Windbase projects to generate more data for his research. Finally, he ended up with a sufficiently large data set to create his machine learning models.
Shen developed multiple models, which he optimized for performance, development time and robustness. He set up a framework to do this, which allowed him to quickly come up with results and make adaptions if necessary. The research results were very promising, even good enough that we could expand the initial scope of Shen’s research and add more detailed information of the FEA outcomes. Shen surprised us all with the results, which was also appreciated by his UT supervisors, resulting in a high grade for his research thesis.
Added value to Windbase
Of course, Shen’s research goal was scientific in nature, but due to the promising results, we immediately expanded on his research. Using his data set and findings, we are now developing new tools which can give our designers direct feedback on the capacity and stiffness of their designs. These tools can be embedded into our current design workflows. Our experts create a design just like they normally would. However, with these tools they don’t have to wait for several days to get feedback on their designs. Instead, they get it immediately when they finish their design and they can even see the impact of small changes in their design. This allows them to optimize their designs in an early phase of the design process, resulting in less expensive designs which are better for the environment. Furthermore, we have started expanding our data set with better and more realistic models. This will allow us to keep improving our designs and services in the future.
Designers have to create a design that meets all the technical requirements, but minimalizes the amount of concrete and reinforcement used. With smart tools, we can provide them with direct feedback, showing which designs pass the requirements and what the material usage (seen in the color scale) of such designs are.