Accurately predicting multiaxial fatigue life of engineering materials is crucial for ensuring structural service safety. Traditional life prediction models mostly rely on the progressive calibration of material-specific fatigue parameters, which seriously limits their applicability across different industrial scenarios. However, purely data-driven models, despite their strong fitting capabilities, commonly suffer from a lack of physical interpretability and a tendency to violate mechanical laws. To address these issues, this work proposes a material parameter-free physics-informed neural network (MPF-PINN) prediction framework. The input features consist of loading parameters directly extracted from testing signals and loading path characteristics obtained through path vectorization. This dual-source feature construction eliminates the dependence on the calibration of material fatigue parameters. Furthermore, macroscopic damage evolution laws are transformed into partial differential inequality constraints embedded within the loss function, ensuring that the predictions conform to fatigue physics from the modeling mechanism. Validation performed on 569 datasets covering 12 alloys and 24 loading paths demonstrates that 91.92 % of the predictions made by the MPF-PINN model fall within the ± 1.5 × error band. Comparative analysis with an artificial neural network (ANN) sharing the same architecture reveals that the proposed framework effectively mitigates overfitting, enhances generalization stability, and avoids fundamentally non-physical predictions. This work provides a new paradigm for high-precision multiaxial fatigue life prediction unified across alloy systems under small-sample conditions.
Additive manufacturing (AM) of high-strength steels, such as 18Ni300 maraging steel, has attracted considerable attention in aerospace and tooling applications due to its excellent mechanical properties and design flexibility. However, the durability of AM components is significantly affected by both cyclic loading and environmental factors, particularly corrosion. This study investigates the combined effect of heat treatment and corrosive ageing on fatigue performance and hardness of additively manufactured 18Ni300 specimens. Corrosion tests were carried out in an ageing chamber according to ISO 9227 using the acetic acid salt spray (AASS) method for 7 days, at a temperature of 35 ± 2° C. Half of the specimens were exposed to these conditions, while the remaining samples served as a reference group. Hardness measurements (Vickers method) were performed to evaluate the influence of heat treatment and corrosion on surface properties. All specimens were subjected to cyclic loading to determine fatigue behaviour. Fractographic analysis was carried out to identify crack initiation and propagation mechanisms. The results demonstrate the influence of the corrosive environment on fatigue life and hardness, providing information on the degradation mechanisms and contributing to the optimization of post-processing strategies for AM maraging steels under demanding service conditions.
This study examines the influences of milling-induced surface topography and residual stresses on the fretting fatigue behaviour of high-strength steel 34CrNiMo6 + QT (AISI 4340) under flat–flat contact conditions. Various milling parameters (feed rate, cutting depth, and tool configuration) were systematically varied to characterize their influence on surface roughness and residual stress states, which ranged from tensile to compressive depending on tool stiffness and cutting conditions. Derived milling configurations were subsequently employed to produce specimens for plain fatigue and fretting fatigue experiments under actively controlled slip amplitude and defined contact pressure. The resulting S–N curves revealed a 78% reduction in fretting fatigue strength compared to plain fatigue attributable to tribological damage mechanisms. While lower surface roughness promotes the crack growth in the inclined section, compressive residual stresses appear to moderately improve the fretting fatigue limit by enhancing crack arrest, as indicated by marker load tests. Moreover, the breadth of the surface power spectrum, expressed by the Nayak parameter, exerts only a minor influence on the normalized number of load cycles. Instead, the fretting fatigue life is primarily affected by adhesive wear and mechanical interlocking influencing the crack initiation. Adhesion formation is largely governed by macroscopic surface waviness and mostly independent of residual stresses. These results indicate that the residual stresses mainly influence the fretting fatigue limit through crack arrest while the surface topography modifies the level and depth affected by contact stresses and primarily governs the specimen lifetime in the HCF regime.
This study investigates the near-threshold fatigue crack growth behaviour of heat-affected zone (HAZ) microstructures in welded structural steels with nominal yield strength of 235MPa and 355MPa. Quantifying the influence of these microstructures is important for reliable assessment of the remaining fatigue life of welded structures. The microstructures of the HAZ subregions were experimentally simulated on material samples representative for bridge structures with a Gleeble thermal-mechanical simulator after which compact tension specimens were extracted. The fatigue crack growth tests were performed with ΔK-decreasing procedure at load ratios of R=0.1 and R=0.5. The crack length was monitored using the alternating current potential drop technique. At R=0.1, the simulated HAZ subregions of both steels exhibited lower crack growth rates and higher long crack threshold stress intensity factors, than their respective base metals. This increased resistance was attributed primarily to enhanced roughness-induced crack closure. At R=0.5, no significant differences were observed between the simulated HAZ subregions and the corresponding base metals, consistent with the suppression of crack closure effects at higher load ratios.
Non-proportional multiaxial stress states deserve more explicit consideration in vibration fatigue, as they naturally arise in this context. Under broadband vibration loading, modal superposition gives rise to phase-lagged stress interaction, rotating stress directions, and complex in-plane shear trajectories, which in vibration fatigue are neither adequately described in their statistical characteristics nor consistently reflected in corresponding fatigue criteria. This paper develops a phase-aware statistical framework for the description of such stress states under vibration loading. A central contribution is the introduction of a complex-valued covariance matrix, which extends the classical second-order description by the full phase-bearing characteristics between stress components. Using candidate-plane stresses as application context, the framework enables the characterization of the correlation structure and trajectory class of non-proportional stress states. To demonstrate its practical relevance, the Findley criterion is adopted as a representative critical-plane criterion for random-vibration and high-cycle-fatigue applications. Based on the proposed descriptors, an efficient screening concept is formulated to identify Findley-relevant plane orientations in large-scale FE models, extending the popular maximum-variance criterion toward non-proportional stress states. Calibration is performed by Monte-Carlo simulation and followed by validation on a broadband-excited multiaxial FE model. The paper contributes to raising awareness of the importance of non-proportional stress states in vibration fatigue, to improving their statistical description, and to implementing these advances in practical fatigue assessment.
The application of Haynes 230 (HA230) by Laser Powder Bed Fusion (LPBF) is limited by its high susceptibility to process-induced cracking. TiB2 addition offers a promising route to suppress cracking; however, its effect on the fatigue behaviour of LPBF-processed HA230 remains unclear. In this work, an LPBF-processed HA230 superalloy modified with 1.5 wt.% TiB2 was investigated in terms of its microstructure, tensile properties, and Low-Cycle Fatigue (LCF) behaviour at 23 °C and 850 °C. The addition of 1.5 wt.% TiB2 effectively suppressed LPBF-related cracking and increased the yield strength by approximately 30 % at 23 °C and 20 % at 850 °C while maintaining high ductility. The enhanced strength is attributed to the transformation of TiB2 during LPBF, which resulted in the formation of M6(CB) carboborides and a fine dispersion of Ti-rich and La-rich nanoparticles. These nanoparticles impeded dislocation motion and contributed to strengthening. Under LCF loading, HA230 exhibited pronounced cyclic hardening at 23 °C, whereas cyclic softening predominated at 850 °C. Fatigue crack propagation was predominantly transgranular at both temperatures, although the damage mechanisms depended on the testing condition. At 23 °C, failure was mainly associated with brittle cracking of M6(CB) particles, whereas at 850 °C, oxidation-assisted surface crack initiation and propagation dominated. This study provides the first comprehensive correlation between the microstructure, tensile response, and LCF behaviour of TiB2-modified LPBF-processed HA230, demonstrating that controlled TiB2 addition can simultaneously improve LPBF processability and mechanical performance.
Laser Powder Bed Fusion (L-PBF) has become a viable approach for manufacturing near-net shape components from a variety of metals. Nevertheless, the application of L-PBF to stress-critical components is hindered by variability in the fatigue properties and contributing process defects. Here, we describe a round robin program performed to i) characterize variability in the finite life fatigue properties of Grade 5 Ti6Al4V produced by L-PBF, and ii) elucidate the contributing defects through quantitative fractography. Three partners produced twelve builds with identical commercial machines enabling the fatigue response to be defined spatially across the build volume. The metal was subjected to either a stress relief (SR) heat treatment or low-temperature, high-pressure hot isostatic pressing (HIP), followed by cyclic axial loading to failure at a single stress level. Results showed that the root cause of failure in the SR condition was a nearly balanced contribution of lack of fusion (LOF) voids and foreign object debris (FOD). Using radial root cause diagrams (RRCD)s, significant spatial variation in the contributions of LOF to the fatigue life was identified, which ranged from approximately 25 % to nearly 75 % of the origins. The HIP treatment successfully eliminated LOF as a contributing root cause and reduced the overall degree of variability in the fatigue response. Through the RRCDs we show that there is an underlying community of anomalies in L-PBF metal after successful HIP treatment, which may require new approaches to quality control.
This study develops an event-driven peridynamic framework for modelling rolling contact fatigue (RCF)–wear competition at the wheel–rail interface. An energy-based bond fatigue model and the USFD wear law are treated as modular sub-models and coupled through a unified cycle-advancement strategy. At each iteration, the minimum of the fatigue-related, wear-related, and remaining-cycle increments updates fatigue states, wear depths, and surface geometry on a common cycle scale without simulating every rolling cycle. A particle-size sensitivity analysis shows that a minimum spacing of 0.10 mm provides a practical compromise among result stability, crack-path resolution, and computational cost. The framework is applied to a two-dimensional wheel–rail model to examine the effects of wheel load, creepage, and friction coefficient. Under the selected parameters and mild-wear conditions, increasing wheel load or friction coefficient promotes crack propagation and wear accumulation, with the response tending towards RCF dominance. Creepage produces a non-monotonic crack response: increasing creepage initially promotes crack propagation, whereas further increases after tangential-force saturation mainly enhance sliding and material removal, strengthening the relative influence of wear. The no-wear comparison shows that wear-induced material removal reduces the retained crack length and modifies the crack path. The fatigue-rate sensitivity analysis shows that the quantitative crack response and the occurrence of a complete fatigue-to-wear transition remain parameter-dependent. The framework provides a unified method for investigating coupled fatigue crack evolution and wear-induced surface removal under repeated wheel–rail contact.
The high-cycle fatigue (HCF) behavior and slip-induced failure mechanism of dissimilar Ti60/Ti17 linear friction welded (LFW) butt joints were investigated under different stress ratios (R = −1, 0.1, 0.3, and 0.5). The fatigue strength at 107 cycles decreased monotonically from 176.1 MPa to 79.9 MPa with increasing stress ratio. Despite the strong microstructural heterogeneity of the LFW joint, fatigue failure was mainly localized in the weaker Ti60 base metal, identifying it as the fatigue-critical region. With increasing stress ratio, the dominant crack-initiation mode shifted from surface to subsurface facets, both associated with basal-slip-induced transgranular cleavage of primary α grains. More importantly, a boundary protection mechanism involving slip transfer and cyclic dynamic recovery was identified, which reduces the tendency for boundary cracking and promotes cyclic plasticity localization within basal slip bands. Based on the quantified facet characteristics, a fatigue life prediction framework was established by incorporating microstructural facet size and mean stress effects into an analytical fracture mechanics model. Furthermore, a physics-guided deep neural network was developed by integrating fracture-mechanics constraints with data-driven learning, achieving reliable life prediction for the independent validation subset with R2 = 0.96. These findings provide a mechanistic basis for fatigue assessment and life prediction of dissimilar titanium alloy LFW joints under asymmetric cyclic loading.
Fatigue and durability assessment of wooden structural components remains a major scientific challenge owing to the intrinsic complexity of wood behavior, particularly its pronounced anisotropy, mixed-mode crack propagation mechanisms, viscoelastic response, and strong sensitivity to environmental variations. This paper delves into a numerical framework combining fracture and fatigue modeling under mixed-mode loading, while simultaneously accounting for anisotropy, viscoelasticity, and moisture variations, in order to predict the service life of wood structures subjected to cyclic hygro-mechano-viscoelastic loading conditions. The energy release rate, which constitutes the key driving parameter of the analysis, is evaluated through the invariant integral Aθvisco, enabling the treatment of viscoelastic mixed-mode fracture under variable environmental conditions while ensuring a rigorous separation of elementary modal contributions. This energy release rate is subsequently introduced into the Paris law to estimate fatigue life through crack growth integration. Several cyclic loading scenarios are investigated, including purely mechanical cycles, coupled hygromechanical loading, as well as viscoelastic and mechano-sorptive solicitations. A comparative analysis of the predicted fatigue lives highlights the influence of hydric cycles on wood durability, particularly when coupled with delayed deformation mechanisms. The results emphasize the necessity of developing predictive fracture-fatigue models incorporating time-dependent and hygroscopic phenomena in order to improve durability assessment, structural diagnosis, and long-term monitoring of timber structures.
Impact pits induce local stress concentration and plastic deformation, leading to fatigue life degradation and challenges in structural reliability assessment. This study proposes an interpretable physics-informed Bayesian neural network (PIBNN) for probabilistic fatigue life prediction of 2024 aluminum alloy with impact pits. Pit-sensitive features, including pit depth, stress concentration factor, and projected plastic area Ap are extracted from profilometry and elastoplastic finite element analysis. Fatigue-informed monotonic constraints are embedded in the loss function to ensure physically consistent predictions. Under separate training, the constrained model achieves a median life R2 of up to 0.91 and produces quantile life curves that better envelope the experimental data. Multi-depth mixed training further improves the 400 μm group, achieving an R2 of 0.968 and 100% coverage of the 90% prediction interval. SHAP analysis shows that the contribution of Ap increases from 7.5% to 16.0%, suggesting increased model sensitivity to local plastic deformation near the impact pit.
The prediction of residual strength of composite laminates under fatigue loading is particularly complex. Most existing prediction methods are based on phenomenological models while ignoring micro fatigue damage mechanisms. To address this, this study proposes a residual strength predicting method based on analysis on three types of micro damage: transverse matrix cracks, matrix crack induced delamination, and fiber fracture. They are explicitly modeled using finite element analysis accounting for the coupled effects of multiple damage modes coexisting in one single laminate. Damage control parameters corresponding to each mode of micro damage are defined to quantitatively characterize the damage level. The relation between the damage control parameters and both the residual stiffness and strength are calculated via a Python script utilizing ABAQUS, which is subsequently used to train a neural network to establish a quantitative correlation between the residual strength and stiffness. Finally, validation with experimental data shows that predicted residual strength curves under different layups and fatigue loading conditions are in good agreement, proving the effectiveness of the proposed method.
