Machine Learning-Driven Quality Control in Additive Manufacturing of Aerospace Components

Abstract

The use of additive manufacturing (AM) in aerospace to manufacture lightweight, high-performance components is on the rise, but to ensure uniform quality, process variability, microstructural flaws, and real-time monitoring must be considered. Conventional endpoint inspection systems, such as X-ray computed tomography and destructive testing, can add up to 30% to production costs and lengthen certification schedules. This paper presents a machine learning (ML)-based quality control system for aerospace-grade metal AM parts that combines in-situ sensor data, thermal images, and layer-by-layer process control parameters to enable real-time defect prediction and process optimization. The dataset analyzed comprised 12,500 build layers from laser powder bed fusion (LPBF) systems. It included more than 40 process variables, such as laser power, scan speed, melt pool temperature, and acoustic emissions. The supervised learning models, such as Random Forest, Support Vector Machine, and Gradient Boosting, were trained to classify porosity and lack-of-fusion defects. The optimized Gradient Boosting model achieved 94.6% classification, 92.3% precision, and 95.1% recall, which was 38% less than the baseline threshold-based monitoring. The predictive framework implementation reduced defects to 3.2-8.7% in validation builds, and reduced the inspection cost by 21%. The statistical analysis also showed a strong correlation between melt-pool temperature variance and porosity (R2 = 0.87). The suggested methodology facilitates the identification of anomalies at an early stage in a single layer in 0.8 seconds, and it supports closed-loop control. The findings indicate that ML-based quality management has enabled high reliability, reduced waste, and enabled certification preparation in aerospace systems. The framework offers a scalable approach to intelligent, data-driven safety-relevant aerospace manufacturing.