Argonne using machine learning to optimize designing and manufacturing of new products; HD engine simulation presented at WCX
16 April 2018
Researchers at the US Department of Energy’s (DOE) Argonne National Laboratory are developing advanced machine learning techniques with high-performance computing (HPC) to help organizations reduce design time from months to days and slash development costs. In the latest example of the work, the Argonne team, with colleagues from Convergent Science and Aramco Research, presented a paper at the WCX 16 SAE World Congress Experience on using a machine learning genetic algorithm (ML-GA) on a heavy-duty engine (Cummins ISX15) simulation.
Machine learning is a type of artificial intelligence that trains computers to discover hidden patterns in data to make novel predictions without being explicitly programmed. This technology can be applied to manufacturing to quickly find the best design for a product or the most efficient production process.
The traditional approach to optimizing the design of a new product involves much experimental testing and evaluation of many prototypes. As the volume and complexity of data derived from these tests increase, industry relies more and more on high-fidelity computer models that virtually represent real-world devices and processes.
These models take in certain values corresponding to controlled aspects of the manufacturing process, for example, fuel pressure, when the fuel is injected into the case of an engine. By using data drawn from physical experiments, the model can determine how well the set of inputs would achieve the desired outputs, such as efficiency and cost-effectiveness.
While they are an improvement over costly investments in physical development and testing, high-fidelity models take a long time to run due to their computationally intensive nature.
Argonne’s solution is to augment high-fidelity modeling with machine learning to accelerate the process significantly, while maintaining the reliability of the data. A job that might take hours to run using high-fidelity modeling takes milliseconds when augmented by machine learning.
In the study presented at WCX, the team used a novel ML-GA to perform numerical optimization of a gasoline compression ignition (GCI) engine operating with a low-octane gasoline-like fuel. A total of 2,048 samples were randomly generated using Monte Carlo sampling and 3D CFD simulations for these samples were performed concurrently on Argonne’s Mira supercomputer.
The samples were distributed over nine input parameters related to the fuel injector design, injection strategy, initial in-cylinder chamber pressure and temperature, and swirl flow.
The team formulated a merit function consisting of 5 targets related to engine performance and emissions. The overall output of the ML model was the merit value. Subsequently, a stochastic global optimization GA was used with the ML model to optimize the merit value.
… the overall turnaround time was (at least) 75% lower with the ML-GA approach, as the training data was generated from concurrent CFD simulations and employing the ML model as the objective function significantly accelerated the GA optimization. This study demonstrates the potential of ML-GA and high-performance computing (HPC) to reduce the number of CFD simulations to be performed for optimization problems without loss in accuracy, thereby providing significant cost savings compared to traditional approaches.
—Moiz et al.
Machine learning converts the very complex physical processes represented by the virtual model into a compact computational process that can be run in much less time. It’s similar to how biologists study fruit flies instead of humans. The flies share significant characteristics with humans, but they can generate and evolve much faster.
—Janardhan Kodavasal, a mechanical engineer in Argonne’s Energy Systems division, who heads the initiative
As demand for machine learning continues to increase, Argonne scientists will continue to expand its competencies. Techniques such as active learning will allow machine learning models to interact with high-fidelity models to improve accuracy and efficiency as the models provide data, and real-time optimization that will help guide manufacturing processes as they happen.
Resources
Moiz, A., Pal, P., Probst, D., Pei, Y. et al. (2018) “A Machine Learning-Genetic Algorithm (ML-GA) Approach for Rapid Optimization Using High-Performance Computing,” SAE Technical Paper 2018-01-0190 doi: 10.4271/2018-01-0190.
Comments