A team of researchers from the University of Michigan and Northeastern University has developed an AI-powered program capable of maximizing the efficiency of any given design. It is indicated that this makes it possible to significantly reduce the time for finding the optimal solution.
The new online learning algorithm is based on a deep neural network and allows you to calculate the optimal parameters depending on the given data. According to the authors of the project, their development makes it possible to solve large multidimensional optimization problems and surpasses any existing analogs, which was confirmed experimentally. The program is able to calculate the design parameters that provide the highest load capacity at the lowest cost.
The researchers tested their algorithm in four completely different scenarios:
- Minimize compliance by designing structures with maximum stiffness to support a given load
- Fluid optimization by designing the shape of fluid channels to minimize pressure loss
- Creating a Shape to Improve Heat Transfer
- Minimization of the required material to create load-bearing structures
The report of the researchers states that the application of their algorithm can reduce the computation time by about 100 to 100,000 times compared to the direct use of heuristic methods.