Neural optimization machine: a neural network approach for optimization and its application in additive manufacturing...

Area of Science:

• Materials Science
• Computational Science
• Engineering

Background:

• Neural networks (NNs) are increasingly utilized in design for objective functions and constraints, necessitating NN model optimization.
• Optimization of NN models concerning design variables is crucial for advanced engineering applications.

Purpose of the Study:

• To propose a Neural Optimization Machine (NOM) for constrained single/multi-objective optimization.
• To integrate NN optimization with additive manufacturing (AM) process-property models and physics-guided machine learning.

Main Methods:

• Designing NN architecture, activation, and loss functions for optimization.
• Utilizing the NN's backpropagation algorithm for seamless integration with AM models.
• Applying a physics-guided NN for fatigue performance prediction and process parameter optimization in AM under uncertainties.

Main Results:

• NOM effectively handles constrained optimization problems, showing minimal computational cost increase with higher dimensional design variables.
• Demonstrated successful application in optimizing AM processing parameters for enhanced mechanical fatigue properties.
• Achieved physics-compatible process design by integrating physics/knowledge with data-driven models.

Conclusions:

• The NOM offers a unified framework for NN training and optimization, eliminating the need for separate optimization processes.
• This methodology enables the design of AM processes with optimized mechanical fatigue properties by integrating physics-guided learning and data-driven models.
• The study highlights a novel approach for constrained process optimization in AM, enhancing structural integrity applications.