Ehsan Hosseini

Metal additive manufacturing, particularly laser powder bed fusion (LPBF), offers major advantages in design flexibility and near-net-shape production. However, the steep thermal gradients inherent to the process generate high residual stresses that can lead to distortion or cracking, thereby compromising part integrity. Conventional optimisation of process parameters through trial-and-error experimentation is costly and time-consuming, underscoring the need for reliable modelling strategies. In this study, two part-scale simulation approaches, the inherent strain method and the block dump method, were systematically evaluated for residual stress prediction in LPBF of Hastelloy X. Representative finite element implementations were developed and benchmarked against experimental observations, including cantilever deflections and residual stress fields obtained from neutron diffraction. Both methods captured the experimental trends with good overall consistency. The block dump method achieved slightly higher accuracy without requiring calibration, but at the expense of increased computational cost and reliance on a more comprehensive set of material input data, including temperature-dependent thermal and mechanical material properties. Sensitivity analyses further showed that predictions from the inherent strain method are highly dependent on the calibration of the inherent strain tensor, whereas the block dump method is more robust with respect to modelling assumptions. Finally, the results of this study indicate that while both strategies offer practical predictive capability, the block dump method provides better generalisation capability for residual stress prediction in LPBF components.

Advanced constitutive models are essential for accurately assessing the integrity of mechanical components, particularly those operating under complex thermomechanical loading scenarios. However, calibrating such models typically involves determining a large number of parameters through inverse analysis, a computationally demanding task due to the need for repeated numerical integration of the governing equations. Scientific machine learning offers promising avenues for reducing this burden, with Physics-Informed Neural Networks (PINNs) emerging as a particularly relevant tool. This study develops a parametric PINNs framework that serves as a computationally efficient surrogate for advanced constitutive models and enables their integration within inverse analysis for calibration—examined here for the Chaboche viscoplastic model across a wide temperature range. We outline key challenges in setting up and training PINNs, and adopt mitigation strategies by incorporating domain knowledge from mechanical engineering. Once trained, the PINNs enable rapid one-shot predictions of stress–strain–time responses and offer automatic gradient evaluation for inverse analysis, alongside acceleration through GPU-based computation. Benchmarking against conventional calibration methods demonstrates that the proposed PINNs-based inverse analysis framework reduces computational time by over three orders of magnitude. This significant improvement paves the way for the practical integration of advanced calibration strategies, particularly for the probabilistic calibration of constitutive models and uncertainty quantification through approaches such as Bayesian inference.

The eigenstrain method assumes that residual stresses originate from an underlying inelastic strain field and has demonstrated considerable success in predicting residual stresses in welded components. Several attempts have been made to extend this approach to additive manufacturing (AM). However, the reliability of their predictions has remained limited. This is mainly because, while a constant eigenstrain field can often be assumed along a weld line, deposited under quasi-steady-state thermomechanical conditions, AM does not hold such characteristic behaviour, and instead, the eigenstrain field in AM parts is inherently heterogeneous. Its local magnitude depends strongly on both global and local geometric features of the component, making the assumption of a constant inherent strain field physically invalid. This study proposes addressing this challenge by learning the heterogeneous eigenstrain field using a recently developed Geometry-Informed Neural Operator Transformer architecture, GINOT, with training data generated from thermomechanical simulations. Once trained, the model would predict the spatially varying eigenstrain field dependent on geometry, enabling residual stress prediction through a one-shot linear-elastic finite element analysis for the entire component. Within the present proof-of-concept study, the proposed framewok is applied to predict residual stresses in LPBF-processed Hastelloy X, where GINOT is trained on eigenstrain fields extracted from the socalled 'block dump' thermomechanical simulations. The results provided preliminary evidence that machine-learning-assisted eigenstrain modelling can achieve residual stress estimation in LPBF parts at a computational cost that is only a fraction of that required by existing thermomechanical or inherent strain simulation techniques. The proposed framework offers a promising pathway toward fast, geometry-informed residual stress prediction for integration into AM design and optimisation workflows.

Fast and accurate representation of heat transfer in laser powder-bed fusion of metals (PBF-LB/M) is essential for thermo-mechanical analyses. As an example, it benefits the detection of thermal hotspots at the design stage. While traditional physics-based numerical approaches such as the finite element (FE) method are applicable to a wide variety of problems, they are computationally too expensive for PBF-LB/M due to the space- and time-discretization requirements. Alternatives are sought to lower the computational burden of modelling this process and make part-scale simulations feasible, with machine learning (ML) techniques leading these efforts due to their exceptional flexibility and efficiency. Recently, there has been a growing trend towards integrating physical insights of the studied phenomena in ML workflows to improve their effectiveness. For the presented work, we hypothesized that the moving laser heat source could be treated as a sequence of pulses such that the solution to various scan patterns could be determined based on the thermal response to a laser pulse. First, a base function represented by a feed-forward neural network (FFNN) was proposed to establish the solution for laser scanning over a wide solid block. Next, inspired by the perturbation theory, a second FFNN was introduced to consider the impact of geometrical features on the temperature profiles as a correction to the base solution. The feasibility of training the pair of FFNNs within the proposed ‘pulse approach’ framework based on a few inexpensive FE simulations, and generalization to larger simulation domains are demonstrated. For a scan pattern not encountered during training, the paired networks are capable of accurately replicating the temperature profiles or history predictions of FE simulations in under one second, showcasing a considerable acceleration by orders of magnitude. The models and scripts used in this study are openly available in https://github.com/HighTempIntegrity/PIAM_Pulse2024.