gpCAM, a software tool for autonomous labs of the future, has been honored with a 2024 R&D 100 Award in the software/services category. Widely recognized as a mark of excellence, the R&D 100 Awards celebrate innovation and the practical application of science and technology across various industries. The winners will receive their award at a ceremony in Palm Springs on November 21, 2024.

gpCAM: software for autonomous labs of the future

gpCAM is a software tool for artificial-intelligence (AI)-driven autonomous decision-making. Scientific facilities around the globe can conduct experiments at an ever-increasing rate, furthering discovery in critical areas such as materials research, drug discovery, carbon capture, energy storage, and many more. The complexity and pace of many experiments make it impossible for them to be controlled by humans, leading to underutilization and inefficiencies. Given a small preliminary dataset, a Gaussian process can be used to compute uncertainties for all possible future experiments and then be queried to return the optimal set of new suggestions. gpCAM (“gp” for Gaussian processes, “CAM” for the project it originated from, CAMERA) allows intelligent decision-making for these experiments faster and more accurately than ever before, leading to accelerated scientific discovery at self-driving laboratories of the future.

The development team led by Marcus Noack and James Sethian includes Ron Pandolfi.

These two technologies, which include significant contributions from Berkeley Lab Computing Sciences staff, were named finalists for this year’s R&D 100 Award.

Optimizing combustion systems with Pele

While alternative energy sources are becoming more widely adopted, combustion-based systems are still ever-present – as is the need to make them cleaner and more efficient. Pele, a software suite designed to simulate and analyze combustion processes, can help scientists understand how different fuels may perform in real-world applications without having to build and test each combination physically. Pele can model and map how fuel properties affect engine and turbine performance for both low-speed (low Mach) and high-speed (compressible) fuel-air interactions, accelerating the development of more efficient combustion systems.

The development team, led by Marcus Day (NREL), includes Ann Almgren, John Bell, Andrew Myers, Andrew Nonaka, and Weiqun Zhang (Berkeley Lab).

DeepHyper: Making Machine Learning Tools More Accessible

Machine learning can accelerate research across fields, from anti-cancer drug discovery to designing more efficient fusion reactors – but how many teams have the resources to adopt and apply the technology? DeepHyper is an open-source Python software package that automates the design and development of trustworthy and energy-efficient machine learning models for scientific and engineering applications. DeepHyper uses novel optimization techniques and parallel computing to speed up and simplify the traditionally labor-intensive configuration process for developing new models, making these powerful artificial intelligence tools accessible to teams working with complex and high-volume datasets and those without a dedicated machine learning expert. The software can be run on devices ranging from individual laptops up to supercomputers.

The development team, led by Prasanna Balaprakash (Oak Ridge National Laboratory), includes Stefan Wild (Berkeley Lab); Romain Egele (University of Paris Saclay); and Misha Salim, Thomas D. Uram, Venkat Vishwanath (Argonne National Laboratory).