From 3 hours to 2 minutes: Subaru adopts AI to shorten the road to its next-generation vehicles
- Customer
- Subaru
- Industry
- Automotive
- Application
- Press forming (die face design)
- Outcome · Time to insight
- From 3 hours to 2 minutes
Overview
To shorten development ahead of a wave of new models planned by 2028, Subaru put Neural Concept to the test on press forming, with Japanese partner Cybernet. In a proof of concept, forming predictions that took 3 hours came back in 2 minutes, with accuracy nearly equivalent to conventional CAE. Subaru then decided to adopt the platform, and plans to extend it across its development process.
Challenge
A wave of new models by 2028 leaves no room for slow feedback.
Subaru aims to start production of a diverse range of vehicles by 2028, including next-generation e-BOXER hybrids and battery-electric models. Meeting that goal means shortening development time for every new model.
Subaru already front-loads its engineering, running repeated CAE studies at the design stage to eliminate problems before production preparation. To go faster, it had to cut the effort those studies take. Press-forming analysis shows why: essential to building safe cars, it could take up to 48 hours per analysis, and one to two weeks before results reached the designers.
Solution
Proof first. Then adoption.
With support from Cybernet, Neural Concept’s partner in Japan, Subaru ran a proof of concept on a specific press-forming task: predicting the thickness reduction of the blank material, a key input for die face design. The platform delivered in 2 minutes what had taken 3 hours, with prediction accuracy nearly equivalent to CAE.
What made the difference was geometry. Subaru’s engineers had found earlier AI approaches too weak at understanding 3D shapes for shape-dependent work like press forming. Neural Concept captured shape features accurately. Based on these results, Subaru decided to adopt the platform.
“Neural Concept has a very high capability for capturing 3D shape features, excelling in accurately reading shape features and simplifying feature quantities. […] we found that Neural Concept allows for faster forming predictions than traditional CAE, while also delivering sufficiently accurate results.”
Impact
From 3 hours to 2 minutesForming prediction, in proof of concept
Subaru first presented the work at Cybernet’s Tokyo seminar in October 2023, and announced its adoption in March 2024. The company expects the gain to significantly reduce man-hours across its development process, easing one of the slowest feedback loops in its front-loading approach.
What’s next
Subaru and Cybernet plan to extend Neural Concept’s predictions beyond this first application, toward CAE-level performance, with the ultimate aim of shortening development across the entire vehicle.
Related content
- Cybernet press release (Japanese) SUBARU adopts Neural Concept for press-forming analysis (March 2024)
- Article Subaru, Honda and Neural Concept Showcase AI-Powered Engineering Innovations at Cybernet Seminar Tokyo (October 2023)
- MONOist (Japanese) SUBARU adopts deep-learning AI tool to cut press-forming analysis effort (April 2024)
Read next