Machine learning-driven nonlinear analysis of inclusion effects in aluminium alloys

TL;DR

This study uses machine learning to show how inclusion size in aluminum alloys affects mechanical strength, fatigue life, and corrosion, offering insights for industrial quality control.

Contribution

The novel contribution is a machine learning framework that quantifies inclusion effects and identifies critical thresholds for material performance.

Findings

Inclusion size is the primary factor affecting tensile strength, with a 8 MPa/µm reduction below 5 µm.

Random Forest outperforms Gradient Boosting with an 18 MPa RMSE in strength prediction.

Fatigue life drops significantly beyond 10 µm, and corrosion rates increase exponentially with inclusion size.

Abstract

The impact of inclusions on the properties of aluminum alloys is comprehensively analyzed in this study using machine learning. The analysis indicates that inclusion size is the primary factor influencing mechanical performance, contributing a significant amount to the degradation of tensile strength in comparison to density’s 35% influence, as quantified by SHAP value analysis. Nonlinear regression modeling identifies critical thresholds, resulting in an 8 MPa/µm strength reduction for inclusions below 5 μm and a stabilization at 275 MPa for sizes exceeding 10 μm. Cluster analysis effectively separates material samples into high-strength (325 ± 10 MPa) and low-strength (285 ± 15 MPa) groups. A comparative model evaluation confirms Random Forest’s superior predictive capability, with an 18 MPa RMSE compared to Gradient Boosting’s 22 MPa. The research quantifies substantial property improvements that can be achieved through inclusion control. The strength is increased by 25 MPa when the size is reduced from 10 μm to 5 μm. However, the fatigue Life analysis demonstrates severe degradation beyond 10 μm, with a decline to 0.5 × 106 cycles compared to 1.3 × 10⁶ cycles at 5 μm in comparison. Corrosion behavior is characterized by exponential dependence, with rates increasing from 0.02 mm/yr at 5 μm to 0.055 mm/yr at 15 μm. A robust framework for comprehending inclusion-property relationships and offering actionable quality control parameters for industrial applications, particularly in aerospace and automotive sectors where precise material performance is critical, is provided by the study’s machine learning approach, which combines predictive modeling with advanced visualization techniques.

Figures