Introduction

As on-demand, highly customizable products become increasingly commonplace, specialized industries, including aerospace, naval, energy, and defense, are seeking alternatives to well-established manufacturing processes that offer higher versatility. Over the past few decades, additive manufacturing (AM) has been propelled to the front line of this search. AM (also referred to as “3D printing”) represents a technological revolution, providing designers with the ability to rapidly prototype many different components with previously unthinkable shapes on a single machine. Interestingly, a second, unexpected revolution also arose from the use of AM, namely, the ability to retain new, far-from-equilibrium states of materials in as-built components, often resulting in materials with intrinsically superior properties.

Of the many existing AM techniques, laser powder bed fusion (LPBF), sometimes called selective laser melting (SLM), is best suited when a balance between component size and printing resolution is necessary. Although theoretically possible, very large (> 1 m) or very small components (< 1 mm) are not typically achievable with commercially available equipment. Instead, LPBF excels at printing a few to hundreds of mm-sized parts with a spatial resolution below 100 µm.1,2 While many different metal alloys have been successfully manufactured with LPBF,3 the stainless steel (SS) alloy 316L (316L SS) is among the most intensely investigated. 316L SS exhibits mechanical and thermal properties that are particularly relevant to LPBF. Due to the local melting, scanning the laser beam results in a small volume of the material rapidly cooling within a larger, colder, body. This builds up significant residual stresses, causing low-ductility materials to crack.4,5 In addition, materials with low laser radiation absorption or high thermal conductivity force the use of high-energy lasers or dramatically affect the thermomechanical history upon printing6. Instead, 316L SS offers both high ductility to accommodate plastic deformation10,11,12,13,14,15,16,17,18 resulting in dramatically increased toughness for structural applications in which a high strength-to-weight ratio is crucial. Nevertheless, it remains to be determined if such microstructural features can be likewise beneficial in corrosive environments.

Fig. 1
figure 1

(a) 316L SS composition (wt%) and the role of each element in the conventional material. (b) Literature reports of different SLM 316L SS properties in recent years (source: Web of Science).

Full size image

The passivation of metals and the local breakdown of passivity in corrosive environments (leading to pitting) have been an intense focus of research in corrosion over the past decades.19,20,21 The pitting corrosion resistance of 316L SS, either conventionally produced or additively manufactured, is primarily attributed to the chromium oxide film (Cr2O3) covering the surface. Therefore, any microstructural features, such as precipitates, elemental segregates, grain boundaries, or dislocation structures that could influence the integrity of this passive oxide film,22 can play a role in either improving or compromising the pitting corrosion resistance. Several studies have been dedicated to understanding the role of microstructure on the corrosion and pitting behavior of LPBF 316L SS.22,23,59

The specific impact of porosity on the pitting corrosion behavior of LPBF 316L SS remains contentious. Sanders et al.26 determined that samples containing the largest distribution of pores with diameters > 10 µm exhibited the highest metastable pitting frequencies and the lowest re-passivation potentials. However, these samples still had high pitting potentials compared to the conventional counterpart. This observation indicates that, although pores can serve as sites for native oxide film breakdown and generation of metastable pits, they do not necessarily lower the pitting potential. Instead, other microstructural features are more likely to elicit stable pitting. On the other hand, other studies found a correlation between enlarged pores in post-processed LPBF 316L SS and lower pitting potentials.118 Interestingly, the cell walls appear to offer a much higher corrosion resistance than the cell interiors, even though the enrichment of Cr and Mo in the cell walls is usually less than 2 and 1 wt%, respectively.35,119 For example, our recent work using in situ atomic force microscopy (AFM) in a 2-M HCl solution shows a clear dissolution anisotropy between the cell walls and interiors, with the former being more resistant to dissolution (Fig. 8). This is likely due to a micro-galvanic effect accelerating the rapid formation of the passive oxide film at the cell walls.120 Another beneficial effect of the cellular structure was an increase in the critical pitting temperature in NaCl.121

The characteristics of the grain and sub-grain structures in as-built LPBF 316L SS have been well studied. Processing parameters have been shown to influence these characteristics in terms of distribution and quantity, but not in essential nature; for instance, high- and low-angle grain boundaries, elongated grains, preferential texture, precipitates, dislocations, and trapped solute elements are always present in LPBF 316L SS. Nevertheless, isolating the individual contributions of each feature to the overall corrosion properties is extremely challenging.

Influence of LPBF Feedstock

The composition, shape, and microstructure of the powder feedstock can influence the surface and bulk properties of LPBF 316L parts. During LPBF, the high surface area of the powder particles is exposed to high temperatures, and often > 1000 ppm of oxygen, with exposures depending on processing parameters, build time, and the number of times recycled. Accordingly, feedstock properties are different prior to melting and solidification, potentially influencing the surface properties of the final part.122 In addition, numerous partially melted powder particles remain on the as-built material surfaces, notably on the side and bottom surfaces (Fig. 4). Powder properties have likewise been shown to affect the porosity and surface chemistry of LPBF 316L SS parts.123 Additional factors, such as initial powder preparation method, shape and size, storage condition, recycling process, and spattering, further influence the final part properties.113,123,124,125 As such, understanding and controlling the powder properties during a build or after repeated powder recycling may provide a pathway towards improved corrosion resistance.

316L SS powders are typically prepared via gas or water atomization, which influence porosity, Mn-Si-O inclusion distribution, and the surface chemistry of the powder particles. Specifically, gas atomization has been shown to result in powder porosity that leaves defects filled with inert gas, leading to bubble formation during LPBF.126 Rapid solidification during gas atomization also results in heterogeneity in the powder chemistry, as shown by EDS measurements of 316L powder surfaces.126 In contrast, water atomization processes result in a higher density of Mn-Si-O inclusions in powder feedstocks compared to powders prepared using gas atomization.127 Such inclusions in the powders have been shown to contribute to inclusions in the printed parts.113 Detailed surface-science studies combining SEM imaging, XPS, and TEM have further shown that the atomizing medium influences the surface oxide chemistry of 316L feedstock powders.128 Specifically, the effects of vacuum induction melting, inert gas atomization, conventional nitrogen gas atomization, and water atomization on powder surface oxide chemistry were studied. Both sets of gas-atomized powders contained homogeneous Fe2O3 oxide layers roughly 4 nm in thickness alongside other oxide inclusions. However, fewer oxide inclusions were found in the conventional nitrogen gas-atomized sample, confirming that powder preparation can influence inclusion density in powders and, by extension, in LPBF 316L SS parts.

The importance of powder storage and the in-chamber environment on powder surface chemistry has also been demonstrated. Both intrinsic oxidation of these powders during storage (at 28ºC and 30–50% relative humidity vs. 80% relative humidity) and extrinsic oxidation within the LPBF chamber (varied between 0.0 and 1.0% oxygen) resulted in an increase in inclusion density in the final build.113 This work also suggests that inclusions are oxygen getters, as they also increase in size due to oxidation. In addition, the collection of spatters as a function of extrinsic oxidation indicated large increases in surface oxygen concentration, as measured by Auger electron spectroscopy. As such, control of the storage conditions and the LPBF chamber environment appear to be crucial for tuning the inclusion density and size in the final parts.

The effects of 316L SS feedstock recycling and reuse have been thoroughly studied in comparison to recycling of other alloy feedstocks. Powder recycling alters the powder shape, morphology, microstructure, surface composition, oxide thickness, and formation of δ-ferrite, in addition to increasing bulk oxygen content.113,124 For instance, powder after 30 reuse cycles exhibits increased Si, Cr, and Fe surface oxide content when evaluated with XPS. Powder reuse also results in the formation of single-crystal ferrite and austenite in contrast to the polycrystalline austenite in virgin 316L powders.124 The phase transformation from austenite in the initial powder to 6 vol% of δ-ferrite in the recycled powder after 16 cycles also results in different magnetic behavior, causing powder clustering in the bed.125 These changes in feedstock magnetic behavior induce defects in final parts, including porosity, delamination, war**, and incomplete fusion. Note that δ-ferrite has been observed in spatters after LPBF,124 but is usually absent in the bulk; nevertheless, δ-ferrite in the bulk material was reported after post-process thermal annealing.129 The effects of residual surface δ-ferrite (from recycled powder, adhered spatter, or partially melted powder) on pitting corrosion of as-built LPBF 316L SS parts has yet to be studied; however, δ-ferrite was found to be detrimental to the corrosion resistance of austenite SS when exposed to NaCl solutions.130 Where possible, storage of gas-atomized powders in low-humidity environments and the use of an LPBF chamber with low oxygen content will minimize powder degradation, thereby reducing the number of defects (pores and second phases) formed in the as-built parts.

In conclusion, additively manufactured 316L SS using LPBF exhibits unique microstructural features spanning a wide range of scales (Fig. 3) that may play different roles in improving or degrading corrosion properties. In particular, the signature sub-grain cellular structure is a feature that remains challenging to understand due to presence of crystalline dislocations, trapped solutes, and inclusions at cell boundaries. As a further complication, these microstructural features are interdependent, and can be affected by process parameters, feedstock, and oxygen content in the LPBF chamber. For better understanding of the corrosion mechanisms and certification of the LPBF parts in corrosive environments, quantitative characterization of key local phenomena occurring during passive film breakdown and metastable pitting are necessary. In situ techniques such as high-speed AFM or TEM are critical to this effort; however, their benefit is magnified when combined with advanced simulation methods, which are often better equipped to unravel the individual and collective effects of microstructural features at different length scales.

Modeling Local Corrosion Mechanisms in LPBF

316L SS: a perspective

The microstructural complexity of AM metals and the wide range of possible multiscale interactions present a challenging yet compelling case for new modeling constructs that can directly probe the relevant physical features and mechanisms. The signature sub-grain cellular structure presents particular difficulties due to presence of crystalline dislocations, trapped solutes, and inclusions at cell boundaries. A first, non-trivial step towards modeling the corrosion behavior of LPBF 316L SS consists in digitally reconstructing the AM microstructures. Although experimental techniques for 3D microstructure reconstruction have been successfully applied to AM materials,131 they are resource-intensive and fail to capture the sub-grain cellular structure in the LPBF 316L SS.132 As an alternative, multiscale physics-based modeling that combines powder-scale models for accurate thermal history profile prediction with microstructure models based on the cellular automaton, kinetic Monte Carlo, or phase field methods have become an effective tool. Such physics-based approaches can accurately capture the dynamical evolution of the solidification microstructure and correlate the resulting microstructural features to processing parameters. A detailed discussion of these approaches is outside the scope of this review and has been reviewed in various articles.133,134,135 These simulation tools could potentially be combined with emerging data-driven tools, including machine learning and generative models, which have been applied to complex microstructure generation in other contexts.136

In principle, these digital microstructures can serve as starting microstructures for subsequent corrosion response simulations. However, to date, relatively few simulation studies have probed the specific connection between process/microstructure models and microstructure-dependent corrosion behavior. This section sets out to first review what has been learned from modeling pitting corrosion of conventionally manufactured stainless steel. We then offer a perspective on simulation advances for integrating microstructures and mechanisms specific to LPBF stainless steel 316L, focusing on available and emerging continuum modeling methods.

Simulation Pathways for Pitting in Conventional SS

Pitting is a small-scale, transient, and localized process arising from multiple concurrent physical and chemical factors that are intrinsically convoluted (Fig. 2). A reliable model must correctly capture the dynamic interplay between the operating environment, electrical and chemical potential gradients, and the evolving morphology of the AM microstructures. It must also account for local breakdown of a passive film, the chemical interaction between the electrolyte and the underlying metal, and active mass transport of ions, leading to morphological evolution of the corroding surface. When properly parameterized and calibrated, such a model could predict how pitting could be affected by changes in composition or microstructure, and identify features and conditions with the greatest impact on degradation.

Development of a reliable predictive model for pitting corrosion requires incorporation of both pit nucleation and pit propagation stages, but a computational model that rigorously accounts for both within a single framework is currently lacking. In the absence of a unified framework, a common approach involves analyzing the processes separately, then using the independent relationships to explore their collective impact. In what follows, we review the progress in simulating pit nucleation and propagation in SS using finite element modeling (FEM), phase-field modeling (PF) and the cellular automaton (CA) approach. Future directions and possible extensions of these models to consider AM microstructures are then discussed.

Simulating Pit Nucleation

The earliest stages of localized corrosion breakdown remain the least understood. Among available models to describe localized corrosion, the point defect model (PDM)137,138,139 developed by Macdonald et al. is a particularly elegant approach that accounts for relevant atomic-scale phenomena. The PDM hypothesizes that pit nucleation results from the injection of cation vacancies at the film/solution interface. These cation vacancies transport through the passive film towards the film/metal interface, where they can condense to form voids or nascent pits if the vacancy flux to the interface is faster than vacancy annihilation via a metal oxidation reaction. Moreover, the PDM accounts for the effects of aggressive species (e.g., Cl–) by assuming they alter the kinetics of generation and transport of cation vacancies. The model has correctly predicted the logarithmic dependency of pit initiation potential on Cl- concentration, as well as the effects of certain alloying elements and scan rate. This model has also been employed to predict the inhibitive effect of oxyanions (such as \({\mathrm{NO}}_{3}^{-}\) and \({\mathrm{BO}}_{3}^{-}\)) on the pit breakdown potential of 316L SS by considering the competitive adsorption of aggressive and inhibitive species at O vacancies at the film/electrolyte interface. The breakdown potential is predicted to vary linearly with \(\mathrm{log}\left({\mathrm{X}}^{-}/{\mathrm{Y}}^{-}\right)\), where \({\mathrm{X}}^{-}\) and \({\mathrm{Y}}^{-}\) represent the concentrations of aggressive species and inhibiting oxyanions, respectively. The prediction agrees with experiment results for 316L SS in \({\mathrm{Cl}}^{-}+{\mathrm{NO}}_{3}^{-}\) solutions.140 Notable criticisms of the PDM are that, in its original form, it assumes linear transport kinetics for the diffusion of species within the passive layer, and that it has been used to predict pitting potential, which is related to pit stabilization, not pit initiation as calculated in this model.141

The formation of pits has often been described as a purely stochastic process, with the transition to faster corrosion explained by the formation and stabilization of a small number of pits with high activity.142,143 However, Lunt et al.144 proposed a stochastic reaction–diffusion model that revised this view. Interactions between early formed pits and the adjacent electrode surface were found to develop as regions with enhanced or suppressed pitting susceptibility (dictated by the buildup of aggressive species and potential gradients upon growth of early-formed pits).145,146,147 Numerical simulations have also been employed to analyze the spatiotemporal dynamics of this process.141 The results demonstrate that the onset of pitting corrosion is a cooperative critical process that proceeds according to a chain reaction. These models assume that active pits concentrate aggressive ions that weaken the protective oxide layer, thereby enhancing subsequent nucleation rates of new metastable pits.

Fluctuations in transient current embody all the critical characteristics of pitting, including initiation, temporary growth, and cessation of growth due to re-passivation. A steadily increasing current signals the formation of a stable pit with propagating growth. As such, it has been proposed that, by introducing many more metastable pits than stable ones, one can predict the formation of stable pits.148 In particular, by assuming that the overall probability of pitting encompasses the probability of initiation and that of pit maintenance and propagation, the nucleation frequency of stable pits \(\Lambda \) becomes proportional to that of metastable pits \(\lambda \), as follows149:

$$\Lambda =\lambda \mathrm{exp}(-\mu {\tau }_{c})$$

where \(\mu \) is the probability of re-passivation and \({\tau }_{c}\) is the critical age from mestable pits to survive in order to become stable pits. In practice, these factors that control a possible transition from metastable to stable pitting are highly specific to the material and difficult to parameterize; however, they may benefit from further investigations using high-fidelity modeling.

Simulating Pit Growth

Many numerical models focus solely on the stable growth stage of pitting corrosion. Along these lines, comprehensive reviews exist on modeling and simulation of pitting processes in conventionally manufactured metallic alloys.150,151,152,153 Computational models for pitting corrosion can be classified into two categories based on how they handle the evolution of the corrosion front during pit growth: non-autonomous models and autonomous models.150 The former treats the mass transport kinetics in the electrolyte and the chemical reaction kinetics at the corroding front separately. Solving the mass transport kinetics problem in the electrolyte requires definition of appropriate boundary conditions to account for the effective species concentration or flux (e.g., a constant-flux condition at the metal/electrolyte interface can represent steady-state metal dissolution and pit morphology evolution). FEM fall in this category of non-autonomous models. By contrast, an autonomous model either describes the dissolution/transport kinetics together with the process of pit migration (e.g., peridynamic (PD)154 or phase-field (PF) models155,156,157,158,159,160,169

Fig. 10
figure 10

(a) CA model for pitting that combines electrochemical simulation and mechanical analysis, (b) CA-predicted evolution of the pit front from an initially passivated metal. Adapted from Ref. 166 under terms and conditions provided by Taylor & Francis and Copyright Clearance Center.

Full size image

First applied to generalized corrosion, CA has since been frequently employed to study pitting corrosion. Li et al.190 considered the precipitation of solvated metal cations into a salt film by adding a species that locally decreases the pH of the solution. Wang et al.191 adopted the salt film species in their model and further considered mechanical effects by coupling FEM to the CA model. In this formulation, the corrosion probability is described as a function of the local stress state (see Fig. 10a). Similarly, intergranular corrosion has been studied by incorporating a location-dependent dissolution probability.168,192 The CA method has also been used to investigate the combined effect of passive film breakdown and re-passivation on metastable pits in sputtered nanocrystalline stainless steel.164 The combined effect was found to inhibit the growth of metastable pits, but with varying individual impacts on pit growth. Pitting corrosion has also been studied in 3D using a probabilistic CA model that couples spatially separated anodic and cathodic reactions to local electrolyte properties, including pH166 (Fig. 10b). The same model has also been employed to study generalized corrosion165 and occluded corrosion cells167

Notably, the capability of CA models to deal with large domains, while using simplified stochastic state-changing rules to account for microscopic heterogeneity, makes them well suited for simulating localized corrosion initiation or passivation/re-passivation processes.193 Moreover, CA involves a relatively simple computational implementation. These two advantages make 3D CA modeling particularly attractive for exploring pit initialization in representative LPBF 316L SS microstructures, which feature complex characteristics that span multiple scales. Furthermore, CA can straightforwardly assess response of pit formation to microstructure variation at modest computational cost. However, a primary drawback of CA is that the time and length scales that dictate the model dimensions are not physical quantities, and thus require careful calibration for particular transition rules and experimental observations. Moreover, the state-transition rules are formulated to represent discrete events that are often difficult to parameterize predictively.

Artificial Neutral Network Method

Pitting corrosion in LPBF 316L SS may depend on a host of variables, from the complex alloy microstructure to the surface properties and environmental factors, each of which evolves dynamically under non-equilibrium conditions. This high-dimensional space makes quantifying specific correlations exceedingly difficult if not impossible. One way to deal with this complexity is to incorporate machine learning approaches such as ANN analysis, which does not rely upon physicochemical models and hence are free from preconceived notions of how the system behaves. Given a sufficiently large and reliable database, ANN methods are effective at uncovering hidden relationships that cannot be discerned by inspection or classical statistical analysis.

It should be noted that a sufficiently large, reliable property database is prerequisite for successful ANN training and application. Obtaining internally consistent data across the wide range of relevant literature is a major difficulty, as is excluding irrelevant outliers to refine the database properly. This is particularly concerning for LPBF materials, which feature microstructural characteristics that are both highly variable and span a range of length scales. One approach for overcoming the challenges of data scatter, complexity, and sparsity was recently demonstrated by Zhu et al.,194,195,196,

Conclusion

The enhanced resistance to pit initiation of LPBF 316L SS in chloride solutions has attracted a great deal of interest. However, probing the underlying pitting mechanisms remains an open challenge. Over the past several years, the number of published papers dedicated to the corrosion properties of the AM material has steadily increased; nevertheless, the complexity of the process-induced microstructures has prevented a complete understanding, particularly for the transient, highly localized pitting process. Much can be learned from comparisons with studies of conventional 316L SS. For instance, the improved pitting potential in the LPBF material is associated with the absence of a MnSi phase. However, distinguishing correlation from causality is often difficult, given the rich diversity of microstructural features unique to LPBF 316L SS. Moreover, it is clear that entirely new mechanisms operate in the LPBF parts. Significant progress has been made in experimentally characterizing AM material porosity, residual stresses, melt pool boundaries, and the grain/sub-grain structures as a function of processing conditions; however, a robust correlation between these microstructural features and the properties of the passive oxide film is still lacking.

While most LPBF 316L SS corrosion studies have been entirely experimental, several continuum modeling approaches have been applied to simulate corrosion in conventional stainless steels, which have led to improved understanding of the mechanisms controlling pitting in chloride solutions. In addition, significant progress has been made toward predicting pitting nucleation, particularly with the aid of experimental characterization methods for model calibration. However, a comprehensive model of pitting corrosion would need to incorporate a wide range of multiscale, multiphysics phenomena. For instance, passivation breakdown initiates as an atomic-scale reaction, but pit propagation involves diffusion and electromigration, which span a range of scales from the microscopic to the macroscopic. In addition, a full description of the pitting corrosion process requires the integration of multiscale models that combine full 3D simulations of multiple-pit growth with an electrochemically accurate pit nucleation model, which necessitates further development. Lastly, for these models to be successfully applied to LPBF 316L SS, it is critical that they adopt a high-fidelity reproduction of a statistically representative volume element, including incorporating hierarchical, multiscale microstructural features such as the sub-grain cellular structure.

In conclusion, the next step towards improved understanding of pitting in LPBF 316L SS involves linking the underlying microstructures to the key properties of the passive oxide layer that ultimately lead to breakdown of passivity. In particular, two high-priority questions remain: first, how much do sub-grain structures affect the local properties of the passive oxide film; and second, what is the preferential pit nucleation site? Answers to these questions have begun to emerge, notably thanks to the use of in situ experimental observations at the atomic and microstructural scales. However, it is clear that a far deeper understanding of local corrosion phenomena in LPBF 316L SS could be achieved through increased availability of integrated experiment–simulation approaches, particularly with the aid of emerging data-driven and machine learning approaches that can leverage feedback between both sets of tools.