Argonne’s ML-GA software technology provides a unique artificial intelligence-driven optimization suite to shrink the industrial design cycle for product development across a variety of markets. The basic principle behind ML-GA is the coupling of machine learning (ML)-based fast-running surrogate models (in place of computationally expensive simulations) with a global optimization technique (in this case, a genetic algorithm (GA)). This in-the-loop ML capability accelerates design iterations within the virtual optimization campaign by an order of magnitude compared to the current state-of-the-art approaches employing simulations directly coupled with the optimizer.
ML-GA exclusively uses an advanced ML algorithm, known as Super Learner, to generate surrogate models of target quantities of interest from simulation data. Within ML-GA, the ML surrogate models (once trained) are utilized to provide fast predictions of the performance of design candidates during each iteration of the GA.
ML-GA software encapsulates all the above features in an automated, self-contained, parallelizable, and portable Python workflow. This makes for integration with any simulation software for virtual design optimization purposes. In addition, ML-GA can be readily run on computing clusters, supercomputers and cloud-based platforms.