Fast Surrogate Modeling using Dimensionality Reduction in Model Inputs and Field Output: Application to Additive...

Area of Science:

• Computational Engineering
• Materials Science
• Data Science

Background:

• Surrogate modeling is crucial for efficient analysis of complex systems.
• Parameter dimension reduction techniques offer potential for computational speed-ups.
• Additive manufacturing processes involve significant stochastic variability.

Purpose of the Study:

• To develop a novel surrogate modeling approach combining principal component analysis (PCA) and active subspace (AS) methodology.
• To achieve computational efficiency by reducing dimensions in both input and output spaces.
• To apply the developed method to analyze variability in residual stress in additively manufactured components.

Main Methods:

• Identification of principal components (PCs) and features in output field data.
• Application of active subspace (AS) methodology to map input variables to output features.
• Development of the Principal Component Active Subspace (PCAS) method for dual dimension reduction.
• Demonstration on a realistic additive manufacturing problem with stochastic inputs.

Main Results:

• The PCAS method effectively reduces dimensionality in both input and output domains.
• The surrogate model enabled efficient uncertainty propagation and identification of stress hotspots.
• Global sensitivity analysis quantified the impact of uncertain inputs on stress variability.

Conclusions:

• The PCAS method offers substantial computational gains for surrogate modeling, particularly in generating training data.
• This approach has enormous potential for advancing control and optimization in additive manufacturing.
• The developed surrogate model provides valuable insights into stress variability and its sources.