Researchers develop machine-learning optimizer to slash product design costs

Brett Hansard, Researchers develop machine-learning optimizer to slash product design costs, Argonne National Laboratory,  NOVEMBER 16, 2020.

Speed up the product design optimization process: 

It employs a novel machine learning technique that helps users focus on how to most efficiently target computational resources. (Machine learning is an application of artificial intelligence that allows systems to automatically learn and improve from experience.)

"ActivO runs the simulations in a very smart way and quickly identifies the parts of the design space we should focus on," explained Pal. "A process that used to take two to three months to give you the optimum design can now be completed within about a week."

Approach: 

According to Owoyele, who is the lead author of the paper with Pal, ActivO is a hybrid algorithm that leverages the strengths of two different machine learning surrogate models to obtain superior performance.

"The machine learning models are designed to work cooperatively. Rather than run simulations that are sampled randomly, one of the models allows us to explore the design space adaptively, which essentially guides us to the regions most likely to contain the global optimum. And the other model looks in those promising regions and performs a local search to identify the exact location of the global optimum."

This approach leverages machine learning surrogates to "explore" and "exploit" the design space in a more balanced and efficient manner than traditional evolutionary techniques used in the industry, like genetic algorithms. As a result, Owoyele said ActivO converges to the global optimum by nearly an order of magnitude faster.

Pal added that ActivO runs in small batches of simulations, making it particularly valuable for industrial users, since they often don't have the computational power to run large ensembles of simulations.

Opeoluwa Owoyele et al. A Novel Active Optimization Approach for Rapid and Efficient Design Space Exploration Using Ensemble Machine Learning, ASME 2019 Internal Combustion Engine Division Fall Technical Conference (2019). DOI: 10.1115/ICEF2019-7237.

In this approach, a hybrid methodology incorporating an explorative weak learner (regularized basis function model) which fits high-level information about the response surface, and an exploitative strong learner (based on committee machine) that fits finer details around promising regions identified by the weak learner, is employed. For each design iteration, an aristocratic approach is used to select a set of nominees, where points that meet a threshold merit value as predicted by the weak learner are selected to be evaluated using expensive function evaluation. In addition to these points, the global optimum as predicted by the strong learner is also evaluated to enable rapid convergence to the actual global optimum once the most promising region has been identified by the optimizer. This methodology is first tested by applying it to the optimization of a two-dimensional multimodal surface. The performance of the new active learning approach is compared with traditional global optimization methods, namely micro-genetic algorithm (µGA) and particle swarm optimization (PSO). It is demonstrated that the new optimizer is able to reach the global optimum much faster, with a significantly fewer number of function evaluations. Subsequently, the new optimizer is also applied to a complex internal combustion (IC) engine combustion optimization case with nine control parameters related to fuel injection, initial thermodynamic conditions, and in-cylinder flow. It is again found that the new approach significantly lowers the number of function evaluations that are needed to reach the optimum design configuration (by up to 80%) when compared to particle swarm and genetic algorithm-based optimization techniques.