My partner Isabel Povoa and I started building this application in the summer of 2014. It was a way for me to scratch my own itch and ultimately led to my career change from mechanical engineering to software development.

The Problem

There has been a big shift in vehicle aerodynamics over the past two decades, where the bulk of the development work has moved from the wind tunnel and track testing to computational fluid dynamics (CFD) simulations. As the simulation software and computing power has matured, it is now possible to do the bulk of the conceptual work digitally before verifying one’s assumptions in physical tests. Today it is common for motorsport or commercial sports car development teams to simulate 5-10 ideas per day.

As with many digital tasks, the limitation quickly becomes the person sitting behind the computer. The speed at which you as an engineer can iterate and test ideas is limited by the time it takes you to perform these tasks in your software tools. This pressure is further compounded by the time spent waiting for results to load and menial repetitive tasks such as setting up simulations and creating frequent reports to communicate ideas.

Complete vehicle simulations require in excess of 500 Million data points each (15-50GB of data). In addition to the core result of pressure and velocity fields, there are a variety of quantities that can be derived from these to help understand the flow field better. Loading this data and deriving these additional properties requires beefy workstations and HPC clusters and takes more than 10 minutes per simulation model. This makes quick investigations tedious and interrupts the train of thought you may have had. It also makes it impossible to investigate alternative ideas and interpretations in meetings when the models are not open.

In response, most design teams now automatically generate a standardised set of images for analysis at the end of a simulation. These images describe the flow field with various visualisation techniques such as surface pressure plots like in the figure below or section images.

The section images show orthogonal planes in 3D space (front, side and top views). The flow field is then described in these sections by showing the pressure, velocity, temperature, energy and other quantities that help the engineer understand how the flow behaves. This is very similar to CT scans used in medicine. If you want a thorough sense of the flow field, you need to have as many sections as possible. This means that each car simulation can easily produce thousands of images.

So we solved the problem of loading the data we want to see, but we’re now stuck with how to query it efficiently. We would need to search through folders of images to compare the same two sections for two (or more) different vehicle simulations.

The Solution

Enter kviz, a lightweight solution to display simulation image datasets. By using web technologies, we had maximum flexibilty in experimenting with the user interface. The goal was a tool that has a shallow learning curve, while also satisfying the needs of power-users that use the tool for many hours every day.

The dominant form of interaction is with the keyboard, with which the user can scroll through the flow field at will and flip between simulations and scalar quatities such as the pressure, velocity and temperatures.

At the same time, it is possible to touch the data. You can get the value of a quatitity at any point in space or flip the orientation of the view at points of interest. User interface “chrome” is minimised with as few buttons as necessary, so that the visualisation/data takes the center stage.

In addition to being an analysis tool, the fast loading times and intuitive interface also make kviz an excellent communication tool. It allows engineering teams to discuss the results and explore alternative interpretations together in a collaborative way, something that was impossible to do on the basis of powerpoint slides.

Because these interactive features are coupled with the small resource footprint, the simulation results become portable. It is possible to take your models with you to the wind tunnel to confirm important features in the flow or investigate inconsistencies between the test and simulation more thoroughly on site.