“It’s like programming a coffeemaker.”
When it comes to manufacturing new lightweight, yet strong components for new passenger jets, scientists are treating the process like trying to brew the most delicious cup of coffee.
By using artificial intelligence (AI) and machine learning, researchers at the U.S. Department of Energy’s (DOE) Argonne National Laboratory are intelligently and automatically selecting the perfect settings for a different kind of hot brew — the process of friction stir welding, a common ingredient needed to manufacture airplane components.
In a new collaboration with GE Research, Edison Welding Institute and GKN Aerospace, Argonne computer scientists are putting the power of the laboratory’s automated machine learning expertise and supercomputers to use. By reducing the number of costly experiments and time-consuming simulations with a new machine learning approach, they can generate accurate models that provide valuable information about the welding process in much less time and at a fraction of the cost.
This approach, called DeepHyper, is a scalable automated machine learning package developed by Argonne computational scientist Prasanna Balaprakash and his colleagues at Argonne. Machine learning is a process by which a computer can train itself to find the best answers to a particular question.
“If you’re trying to brew the best cup of coffee, you can spend several hours fiddling with the many settings on the best machines,” Balaprakash said. “In trying to make airplane parts, we can avoid this by using machine learning, which gives us the ability to learn from a handful of example settings and identify the best one from a set of a billion possible configurations.”
According to Balaprakash, the machine learning algorithm uses a training dataset of various welding conditions and parameters from which airplane part properties can be determined. From this dataset, vastly more possible inputs are instantly analyzed and ranked to determine which give the best possible components.
“Manufacturing airplane parts involves highly complex, sophisticated and expensive machines, and automating their manufacturing can save money and time, and improve safety and efficiency,” Balaprakash said.
Just as someone might prefer their coffee strong and bitter, or light and mellow, scientists who use machine learning need to develop different models that look at many different properties of the welding process, giving different answers to which is best for different properties.
DeepHyper automates the design and development of machine-learning-based predictive models, which often involve expert-driven, trial-and-error processes. Because, in Balaprakash’s words, “no model is an absolute reflection of the truth,” he and his colleagues are not primarily trying to find the single best predictive model and the associated welding condition. Rather, they are generating hundreds of highly accurate models, combining them to assess uncertainties in the predictions, and then seeking to use these more tested predictions in the manufacturing process.
The team’s computationally intensive work is being enabled by supercomputing resources at the Argonne Leadership Computing Facility, a DOE Office of Science user facility.
The partnership between Argonne, GE Research, Edison Welding Institute and GKN Aerospace is funded by a grant from DOE’s Advanced Manufacturing Office. The project is entitled Probabilistic Machine Learning for Rapid Large-Scale and High-Rate Aerostructure Manufacturing.