Metal additive manufacturing promises complex parts without expensive tooling, but the quality of each printed part depends on dozens of laser fusion parameters. Inside the material, randomly shaped grains and pores form, becoming sites for fatigue crack initiation. Neighboring work in this area either builds empirical correlations between printing parameters and fatigue, uses neural networks trained on limited datasets, or applies multiphysics modeling without explicit quantitative uncertainty assessment. This work stands apart: it combines experimental microstructure characterization with physics-based models and extreme value statistics to obtain the fatigue life distribution directly from grain and pore geometry.
Cubes with a side of 1.6 mm were cut from 316L steel cylinders printed on an EOS M290 machine. Grains were characterized by electron backscatter diffraction, and pores by micro-computed tomography. Each cube was divided into 64 subcubes with a side of 0.4 mm. For each subcube, three simulations were run: a linear-elastic finite element calculation for the elastic energy release rate, a graph search for the minimum energy crack path for the surface energy, and crystal plasticity for the fatigue indicator parameter. To speed up the elastic calculation, a two-level scheme was used: first a coarse mesh, then its solution is interpolated as an initial guess for a fine mesh with a block preconditioner. For each subcube, extreme values of the three quantities were taken and distributions were fitted by maximum likelihood, selecting the normal distribution based on Akaike and Bayesian information criteria.
For the recommended printing parameters, the mean fatigue life was about 31,400 cycles, and the mode about 29,300 cycles. The distribution was right-skewed, so planning inspections with an 80% margin below the mean would give an interval 4,200 cycles shorter than using the full probability density. Increasing the mean grain diameter from 60 to 80 micrometers reduced the mean life only slightly: from 31,400 to 29,600 cycles. Pore size had a much stronger effect: at the same volume fraction of pores of 1.5%, the configuration with many small pores (diameters 10–80 micrometers) gave a mean life of about 40,400 cycles, whereas the configuration with fewer large pores (50–80 micrometers) gave only about 33,600 cycles. The analytical probability density derived for the ratio of correlated normal variables agreed well with the kernel density estimate from direct calculations in each subcube.
This work differs from neighboring studies in that it does not merely predict the mean fatigue life or classify a part as acceptable or not, but provides the full probability distribution, accounting for correlations among the three physical quantities computed on the same microstructure. This is the first approach that directly links experimentally measured grains and pores to an analytical formula for the fatigue life distribution, bypassing conservative deterministic estimates. Such a result allows decisions about part acceptance to be based on probabilistic criteria rather than arbitrary safety factors.
The method can be accelerated by replacing crystal plasticity with a surrogate model and porting the elastic calculation to parallel PETSc solvers. The correlation coefficients among the three quantities are currently set by expert judgment; they could be calibrated against experimental fatigue tests on real printed specimens. In the long run, microstructure models that predict grains and pores from printing parameters could be coupled with this framework to assess part reliability before manufacturing.
The results are directly applicable to the qualification of critical parts in aviation, energy, and medical implants, where fatigue reliability determines acceptance. The method is also useful for optimizing printing parameters: it shows which combinations of grain size and porosity yield acceptable life distributions.
The next step is to calibrate the correlations using fatigue test data from printed specimens with known microstructure. Then, microstructure prediction models based on process parameters should be integrated to close the loop from printer settings to life distribution.
In materials science, the question of how the three-dimensional pore morphology and crystallographic texture govern fatigue crack initiation remains open. This work shows that even at the same porosity, the pore size distribution changes the mean life by 20%, highlighting the need to account not only for the total defect volume but also for its statistics. The relationship between grain size and fatigue strength, described by the Hall–Petch relation, appears weak here, indicating the dominant role of defects in printed materials.
This section is addressed to the author of the paper and their colleagues. Neighbouring works were found by meaning-based search across our archive, not by word overlap. We do not review: we show the field around the work.
What the work does
The study links the microstructure of 316L printed steel (grains and pores) to the distribution of fatigue life through a combination of finite element analysis, crack path search, and crystal plasticity, applying extreme value statistics to subcubes. An analytical probability density for the number of cycles to crack initiation is derived, and it is shown that many small pores are safer than a few large ones at the same porosity.
Where its strength is
The strength of this work is the shift from deterministic simulations to a probabilistic description of fatigue that accounts for correlations between physical quantities, enabling inspection planning with greater precision than the traditional safety factor approach.
Where it could go next
Worth trying
Why this may matter
This approach could give manufacturers a tool for probabilistic reliability assessment of printed parts and for optimizing print parameters to meet a specified risk level.
Nearby works in our archive
The archive is sparse around this work: the closest review is only 0.54, against a usual 0.64. It looks like few have drilled nearby.
The crystal plasticity simulation for a single subcube took almost 18 hours on a 28-core processor, whereas the elastic calculation finished in 355 seconds, and the graph search for the crack path in 91 seconds. For 64 subcubes and five cases, the total time for plasticity alone exceeded 240 days of machine time.
In Kim Stanley Robinson's novel 'Aurora', the interstellar ship suffers from fatigue cracks in printed structures that cannot be predicted in advance. This work offers precisely the tool the characters lacked: a probability distribution of fatigue life that allows repairs to be planned before a crack becomes critical.