Predictive Modeling for Additively Manufactured Gyroid Lattices

Data Science & Research Internship
Guide: Dr. Praveen Bidare, Sheffield Hallam University

Machine Learning Workflow for Gyroid Lattices

Project Overview

The optimization of 3D printed lattice structures involves highly nonlinear interactions between various geometric and post-processing parameters. Traditional trial-and-error experimental approaches are impractical for such high-dimensional design spaces. In this research, I developed a robust, data-driven machine learning framework to predict performance metrics and bypass computational bottlenecks, establishing a direct mapping between complex geometric parameters and target outcomes.


Technical Implementation

To capture the complex multiphysics interactions effectively, the modeling framework utilized an ensemble of Deep Learning and Machine Learning strategies tailored to specific variance profiles:

ANN Architecture Schematic
Pearson Correlation Matrix

System Output & Features

The deployed machine learning models demonstrated exceptional predictive fidelity across all targeted metrics. The optimized ANN model utilizing Swish activation achieved a coefficient of determination (R²) of 0.9616 for energy absorption and 0.9374 for compressive stress. Furthermore, the physics-informed Gradient Boosting and SVR models reliably predicted stiffness properties (R² = 0.8509 and 0.8257, respectively). This framework successfully proves that data-driven optimization can reliably replace extensive physical testing in additive manufacturing.