Optimizing automotive manufacturing with Generative AI

GEO, a proprietary Generative AI technique, tied or outperformed other state-of-the-art optimization solvers in 71% of problem configurations.

The Challenge

Global manufacturers like BMW face a difficult optimization problem: how do they schedule their workers to achieve production targets while minimizing idle hours? There are a wide range of possible configurations and many constraints. Different shops have different production rates, and each has their own discrete set of shift schedules. What’s more, manufacturers need to prevent overflows and shortages in the buffers between steps in the manufacturing process.

Our Approach

As part of their membership in MIT’s The Center for Quantum Engineering (CQE), Zapata and BMW Group collaborated to apply generative AI techniques to BMW’s plant scheduling optimization problem. Specifically, we trained a quantum-inspired generative model on the best solutions generated by existing state-of-the-art solvers. The generative model then generated new, previously unconsidered solutions. We call this approach Generator-Enhanced Optimization (GEO), and there are many use cases beyond plant scheduling optimization.

Using our computational workflow platform, Orquestra^®, we ran about one million optimization runs cycling through dozens of different categories of algorithms, problem configurations and optimizer solutions to benchmark their performance against each other. We identified the best algorithm to solve each problem configuration, whether it was GEO or a conventional solver. GEO performed particularly well in configurations with large spaces of possible solutions, outperforming all other solvers in the configuration with the largest solution space. Overall, GEO tied or outperformed other state-of-the-art solvers in 71% of configurations.