Assessment of the Critical Defect in Additive Manufacturing Components through Machine Learning Algorithms

Abstract

The design against fatigue failures of Additively Manufactured (AM) components is a fundamental research topic for industries and universities. The fatigue response of AM parts is driven by manufacturing defects, which contribute to the experimental scatter and are strongly dependent on the process parameters, making the design process rather complex. The most effective design procedure would involve the assessment of the defect population and the defect size distribution directly from the process parameters. However, the number of process parameters is wide and the assessment of a direct relationship between them and the defect population would require an unfeasible number of expensive experimental tests. These multivariate problems can be effectively managed by Machine Learning (ML) algorithms. In this paper, two ML algorithms for assessing the most critical defect in parts produced by means of the Selective Laser Melting (SLM) process are developed. The probability of a defect with a specific size and the location and scale parameters of the statistical distribution of the defect size, assumed to follow a Largest Extreme Value Distribution, are estimated directly from the SLM process parameters. Both approaches have been validated using literature data obtained by testing the AlSi10Mg and the Ti6Al4V alloy, proving their effectiveness and predicting capability.

Review on machine learning techniques for the assessment of the fatigue response of additively manufactured metal parts

“…Single source [21][22][23][24]27,[30][31][32][34][35][36] Multi-source [16][17][18][19][20][28][29][30]33 the model, allowing the user to easily consider and implement confidence level bands in the stress life plots. 24 Nonetheless, in, 28 the mean and standard deviation of N f are estimated by using a PINN layout instead of a GPR, with a properly designed custom loss function that uses probability density function and cumulative distribution function with location parameter μ and scale parameter σ.…”

“…Tridello et al, 18 on the other hand, implemented the statistical nature of post mortem fractography (PMF) defects in the simplest manner, by using a straightforward approach with an FFNN that predicts the probability of occurrence of a specific defect size with the SLM process parameters as input. By assuming that the defect size follows the largest extreme value distribution (LEVD), a second ML algorithm that predicts the location (μ) and scale parameter (σ) of this distribution has been developed.…”

“…The table reports also the main output types for the sake of brevity, as the majority of the papers focus on predicting the fatigue life N f 17,21-23,27,29-36 or its mean and standard deviation N f,μ and N f,σ, 24,28 while the others predict other relevant quantities as the crack propagation rate da/dN, 20 the stress amplitude σ a 16,19 and Gumbel LEVD distribution parameters. 18 For a detailed layout of the inputs and outputs used by each article, the reader is referred to Table 10 and Figure 10.…”

“…To ensure effective computations, with high dimensional input vectors and a database with more than 750 training points coming from 76 datasets, a pyramidal NN structure of 3 layers with 100, 75, and 50 neurons has been used. 18 In fact, the Authors in 35 demonstrated that for the three major AM alloys (AlSi10Mg, Ti6Al4V, SS316L), employing pyramidal network structures, with a decreasing number of neurons from the first to the last hidden layer, allows to satisfactorily predict the fatigue performance of AM alloys. Still in, 17 the PINN approach has shown to be significantly superior to a simple FFNN approach, as the training has been guided with prior knowledge constraints to have strictly decreasing stress life curves, with an acceptable computational cost increase.…”

“…Among various ML techniques, neural networks (NN) are the most famous and widely used. [16][17][18][19] Their customizability allows the user to effectively tailor the NN structure and leverage their ability to model complex relationships through multilayer architectures. K nearest neighbors (KNN) is one of the simplest ML algorithms whose predictions are based upon the majority class or average value of its k nearest neighbors in the feature space.…”