INTEGRATING PARAMETRIC OPTIMIZATION AND MACHINE LEARNING TO IMPROVE DIMENSIONAL ACCURACY IN ADDITIVE MANUFACTURING

Abstract

Additive Manufacturing (AM) continues to transform rapid prototyping and low-volume production, particularly for applications requiring complex geometries and fast iteration cycles. However, its broader adoption in precision engineering remains constrained by dimensional accuracy and mechanical reliability limitations. These challenges are especially critical in producing components such as gears for mechanical and electromagnetic counters, where tight tolerances and functional durability are essential.

This study explores using Artificial Intelligence (AI) to optimize AM-produced precision components, focusing on gear applications. Utilizing vat photopolymerization (VPP) or Stereolithography Apparatus (SLA), technologies widely available at the Philippines' Advanced Manufacturing Center (AMCen), this study evaluates how various print parameters and material choices affect dimensional deviation and mechanical durability.

Artificial neural networks (ANNs) train the data from experimental prints and performance tests. Results show training, validation, and test R² values ranged from 0.75 to 0.86, 0.74 to 0.94, and 0.79 to 0.93. The total R² is consistent across 10 K-folds. The lowest MSE was observed in Folds 4, 6, and 7, aligning with the highest test R² values, with Fold 7 achieving the best performance at an MSE of 9.38 × 10⁻⁴ and an R² of 0.92. The overall model performance remains consistent regardless of data partitioning, indicating model stability and reliability.

The experimental results showed that the optimal parameter setting was achieved with a 0.05 mm layer height, 22.5° angle, 15-minute curing time, and 60 °C curing temperature, with curing temperature and layer height exerting the most decisive influence on performance. The prediction profiler identified 0.075 mm layer height, 22.5° angle, 30-minute curing time, and 60 °C curing temperature as the optimal settings, achieving an S/N ratio of 9.51 with a desirability of 0.8435.