A novel machine learning-based model for predicting the transition fatigue lifetime in piston aluminum alloys

Abstract

The estimation of transition fatigue lifetimes for piston aluminum alloys was carried out using unsupervised machine learning (ML) with the K-means algorithm. For this purpose, an experimental dataset representing standard ISO specimens with piston aluminum alloy material, which was subjected to rotational bending fatigue tests under fully reversed cyclic load conditions, was utilized. Subsequently, the stress and fatigue lifetime data were employed to fit the algorithm of K-means clustering. Then, to enhance the K-means performance, various preprocessing methods and Kernel functions were employed to cluster fatigue lifetime and stress data. Furthermore, following the division of the data into multiple clusters, the middle cluster, which represents fatigue lifetime and stress, was identified as the transition fatigue region, and its center defines the estimated transition fatigue lifetime. Ultimately, the transition fatigue lifetimes were determined using the Coffin–Manson–Basquin equation for piston aluminum alloys and compared to the estimated transition fatigue lifetimes, along with the calculation of relative errors. The obtained results indicated that, among the different models employed in this study, the polynomial Kernel K-means clustering algorithm proved to be the most efficient for clustering data within stress and number of cycles plots (S–N plots). Moreover, employing the K-means algorithm with a polynomial Kernel function and five cluster numbers yielded the most accurate estimation of transition fatigue lifetime for piston aluminum alloys, exhibiting the lowest relative error.

Keywords

• Machine learning
• Transition fatigue lifetime
• Piston aluminum alloys
• Polynomial Kernel K-means