PSI - Issue 46

Nithin Konda et al. / Procedia Structural Integrity 46 (2023) 87–93 Nithin Konda et al. / Structural Integrity Procedia 00 (2019) 000–000

88 2

1. Introduction Metal additive manufacturing of structural components in aerospace and biomedical sectors has seen an exponential growth in recent years because of its wide range of advantages, such as high material utilization, short production cycle, reduced parts in assembly and ease of manufacturing(Liu and Shin 2019; Pegues et al. 2018; Xie et al. 2021) . This innovative process has gained considerable attention in Aerospace industry while saving time and money for aircraft manufacturers. Some of the aerospace components such as spars, ribs and longerons are made of either Aluminium or Titanium based alloys, when operated at high speeds, these structures experience vibrations(Chastand et al. 2018; Fotovvati, Namdari, and Dehghanghadikolaei 2019; Vayssette et al. 2018). α-β Ti-6Al-4V (Ti64) is widely used in biomedical applications where cyclic stresses are predominant in corrosive environment. In addition to these applications, this alloy is widely used in Marine Applications, Chemical Industries, Gas Turbines, Firearm Silencers. Due to the rigorous working conditions, vibrations lead to fatigue induced failure which reduces overall useful life of the deployed parts. As a greater number of industries are opting for additive manufacturing technology to produce Ti6Al4V, it is essential to study fatigue life of this alloy (Cain et al. 2015; Mythreyi et al. 2021). Certain inherent flaws are observed in Additive Manufacturing technique, such as poor bonding defects, spherical porosities, crack initiations from edges. These inherent defects make the components vulnerable to cyclic loading. There are some of the post processing techniques such as shot peening, laser shock peening, and post heat treatment methods which are performed on the AM components to enable the parts to be more efficient, strengthened and long lasting. The literature review of AM fabricated Ti6Al4V has revealed that fatigue tests are performed by altering the different variables such as processing parameters, post processing conditions or testing conditions. The present work focuses estimation of fatigue life of the Ti6Al4V fabricated by L-PBF by varying the processing parameters and a testing condition. This estimation is based on developing Machine Learning models on the available data reported in the literature. The fundamental model for fatigue life is based on S-N curve which provides relationship between the Stress Amplitude and the Life of the component.(Kamble, Raykar, and Jadhav 2020; Raja, Chukka, and Jayaganthan 2020) The data is collected from the S-N curves for different processing parameters used for fabricating Ti6Al4V. This work uses two machine learning (ML) algorithms which are predominantly tree-based models to estimate the Fatigue life and both the algorithms are compared on few metrics to understand which algorithm suits this dataset better. The algorithms that used are Random Forest Algorithm and XG Boosting Algorithm which utilizes two different techniques for splitting a node in a decision tree. The fatigue data of this alloy collected from the literature were used to train the data using these two algorithms in order to explore its predictive capabilities of Fatigue life of L-PBF fabricated Ti6Al4V alloy in the present work. 2. Methodology and Modelling 2.1 Data Collection The porosity percentage was estimated based on processing parameters of L-PBF as discussed by Leiming Du et al. (Du et al. 2021) and mentioned in Table 1. Fatigue experiments were also carried out with the same processing parameter’s combination (Table 1), resulting S-N curves which has utilized well in the current studies. The experiments were performed with 10 different combinations of processing parameters to produces mentioned S-N curves.

Table 1. Processing parameters (Du et al. 2021). AM Parameters

Value 1 Value 2 Value 3

Laser Power [P](W) Layer Thickness [t] (µ) Scan Speed [v] (mm/s)

120

160

200

30

45

60

800

1000

1200 0.13

Hatch Spacing [h] (mm) 0.07

0.1

2.2 Data Pre-processing The collected data has 4 different processing parameters. The stress amplitude during testing is considered as