PROJECT Summary
July 2020 - September 2020
This visualization is used to show the output of the regression run on a hydraulic model in the tuning process. We use a number training cases that are comprised of generated CFD data. The tuning process allows us to “fit” a hydraulic model to better represent the air duct network in reality.
The output of the regression is a set of values that are returned in a table. We then use that table to construct this parity chart in the tuning summary so that the engineer can quickly see if the tuning result is satisfactory
Tools:
Sketch
Miro
InVision
Ideation
User Problem
After the regression is run in the model tuning process, a process engineer needs to view the results of the regression to determine if the model is a good “fit”. If it is not, then the engineer can adjust solver variables and re-run the regression.
Typically, the engineer will review the results in a spreadsheet format and look for large disparities. This can be a manual and time-consuming process depending on the model and test data.
We decided to provide a visual representation of the regression results so that engineers can quickly see these large disparities.
Design
Wireframes
I began designing the product around the process of setting up a model tuning.
The first screen is a placeholder “sketch” that I used while designing the model setup mock-ups. The second is my first iteration of a higher fidelity parity chart.
After a feedback session with the principal engineer I learned of a flaw in my design. The parity chart here shows bursts of dots, but the way that this regression works the dots would be in vertical lines.
Final Screens
This is an example of the final screen that was provided to the development team.
The parity chart here shows the model output as the green trend line. The training cases are each vertical set of dots.
A good “fit” model would have dots that are closely aligned with the trend line. A larger dispersion of the dots signifies a greater different between the training data and the model, indicates a bad “fit.
Next Steps
This visualization has been developed as part of the Hydraulic Model Tuner. At this time, the product is in User Acceptance Testing (UAT).
Some next steps would be to gain feedback for users on the value of this visualization. We decided to limit the amount of interactivity for this visualization for the first iteration, but would likely revisit that idea in the user feedback process.
The principal engineer has requested another visualization that would need to be evaluated for the next iteration. It is an adapted bar graph that shows the deviations between the training data and the target specified in the model set up process.