On 10 June, 2025, our colleague Rebecka Mårtensson presented and defenden her Master thesis “An AI Model for Predicting Differences Between Detailed Physical Spray Paint Simulations and Projected Profiles”.

Abstract

Spray painting is an important step of the manufacturing process in the automotive sector, providing durability and visual appeal. Such processes can accurately be simulated using a multi-physics framework. This, however, is computationally expensive and time consuming. An alternative is a projection-based simulation method, gaining simulation speed at the expense of physical accuracy. It performs well for large flat surfaces but poorly at edges and high curvatures. In this thesis, a Convolutional Neural Network (CNN) with U-Net architecture has been developed and trained on the paint thickness difference between physics-based and projection-based simulations. Adding this difference to the projection simulation allows for both fast simulation time and accurate paint thickness predictions.

Simulations of six car parts were performed and the thickness per time step extracted. The applicator path and object triangulation were passed through a ray-tracing algorithm generating a hit- and depth map between applicator and object at each time step. Projected thicknesses and hit- and depth-maps served as inputs to the CNN. The output was the paint thickness difference as a 2D array, an image from the applicator’s viewpoint. Each of the six simulation results were used once as test data, while the remaining five were training data, resulting in six distinct train-test splits. Additionally, a comparison between front and Back Door parts was performed. Model predictions were evaluated against the true targets using Mean Squared Error (MSE) and visual inspection. The resulting MSEs were compared to those of zero prediction baselines, which gave relative improvements of 25\% to 67\%. Both large flat surfaces and more complicated edge-regions seem to be well predicted by the model.

We congratulate Rebecka and look forward to continue working with her here at FCC!