Problem: To help a global automobile manufacturer minimize design time
Business need: A large North American automobile company approached INSOFE with a goal to reduce their design times. Typically, in an automobile design phase, teams spend multiple years perfecting the design for optimizing a number of factors (noise, vibration, aerodynamic efficiency, fuel efficiency, etc.).
As dummy testing is prohibitively expensive, they take the path of simulation (finite element modeling is commonly used). However, even FEM takes hours and extensive computing. More importantly, FEM does not produce a functional relationship between the inputs and outputs. It takes the inputs and solves differential equations on non-linear surfaces smartly using numerical techniques.
The client wanted to see if we can simulate and predict the FEM outputs using the inputs. This shall save them substantial amounts of design time.
To have INSOFE faculty and data scientists solve your business problems, prep your engineering teams to face the real world complexities, visit here
Approach: We constructed a variety of regression models and were able to predict the simulation results at 99% + accuracy after a certain number of simulations. Neural networks with ridge regularization provided the closest approximation.
Another ask was visualizations. We constructed partial dependency plots and were able to automatically capture the feature that has the highest impact on the outcome. This helped engineers plan the next simulations effectively.
Another interesting outcome was that multiple models built for various factors (noise, vibration, etc.) were fed into a multi-objective optimization engine (we used genetic algorithms) to identify the set of design parameters that yielded the most optimum results.
Outcome: They estimated that this model could reduce the number of simulations by 10% without reducing accuracy.