Högskolan i Skövde

Among the various driving forces involved in the molten pool during keyhole laser welding, the vaporization-induced recoil pressure is the dominant one. This study experimentally measured the process-induced recoil force during laser welding of aluminium and copper. A customized measurement setup was used to measure the specimen displacement caused by the recoil force, which was then determined by means of a finite element (FE) analysis. Furthermore, multi-physics computational fluid dynamics (CFD) models of the laser welding process were developed. After calibration, these models were used to predict the recoil force and its dependence on various process parameters. When only the recoil pressure acting on regions where vaporization occurs was considered, excluding the gaseous phases in the model, the total recoil force was underestimated. To account for that the formed gas contributes to the total recoil force as it rises and exits the keyhole, the total recoil force was calculated based on the predicted net mass flow due to vaporization and condensation. This simplified model showed good agreement between predicted and experimentally measured recoil forces, demonstrating that the observed consistent recoil force with increasing laser power may be due to a corresponding increase in the condensation rate. This highlights the importance of understanding the behaviour of the vaporized gas phase to determine appropriate simplifications and assumptions in laser welding process modelling. The findings of this study support the development and validation of multi-physics process models, further advancing knowledge of relevant modelling approximations.

Low-austenitizing temperature practices resulted in substantial changes in both microstructure and mechanical properties of the fully ferritic as-cast Compacted Graphite Irons (CGI). The austempering processes were accomplished through first austenitizing at 850 °C for 60 min followed by quenching in a salt-bath at 275, 325, and 375 °C for times ranging from 30, 60, 90, and 120 min. In contrast with the austenitizing performed at 900 °C performed on the same material, the microstructure consisted of a notable volume fraction of proeutectoid ferrite, which was not observed under similar austempering temperature and time conditions. Lowering the austenitizing temperature to 850 °C resulted in decreased untransformed austenite. Depending on the austempering conditions, a notable improvement was achieved in both Brinell and Vickers hardness compared to the as-cast CGI. The ausferrite matrix led to remarkable increases in yield strength (YS), ultimate tensile strength (UTS), and a decrease in total elongation to failure. The highest YS and UTS values were achieved for specimens austempered at 275 °C while increasing the austempering temperature decreased both YS and UTS. Furthermore, the results showed that the austempering temperature had a more significant impact on YS and UTS than the austempering time. All austempered CGI specimens exhibited primarily brittle failure attributes, while ferritic CGIs showed a mixed failure mode.

Laser welding of busbars to battery tabs in electric vehicles (EVs) is crucial due to the rapid advancements in electric mobility technology. Ensuring weld quality is paramount, as it depends on factors such as porosity generation, fluid flow in the molten pool during welding, applied laser power, and welding speed. However, conventional laser welding techniques, which primarily focus on adjusting laser parameters along the weld direction, struggle to effectively mitigate porosity formation. While the effect of laser angles along the weld direction has been extensively studied, the effects of off-axis laser angles, i.e., angled in the plane perpendicular to the weld direction, have not yet been explored. This study introduces an innovative approach to laser welding by varying the laser off-axis angle at different laser energy densities to optimize the process specifically for porosity reduction. By implementing a three-dimensional computational fluid dynamics (CFD) model of laser welding of aluminum AA1050, we provide a detailed analysis of the fluid flow and melt pool dimensions while employing different off-axis angles. Our model incorporates multiple reflections, upward vapor pressure, and recoil pressure to explain porosity formation at different laser off-axis angles. The results show that increasing the laser off-axis angle at optimized laser power and welding speed significantly reduces porosity. The numerical analysis indicates a maximum deviation from the experimental melt pool width of 11% at a laser off-axis angle of 4.92° and a minimum error of 2.6% at an off-axis angle of 2.74°. For melt pool depth, the maximum deviation is 7.2% at an off-axis angle of 4.92°, and the minimum difference is 0.5% at an off-axis angle of 7.42°. This study presents a novel methodology for improving laser welding processes by addressing the specific challenge of porosity formation.

Laser beam welding (LBW) involves complex and rapid interactions between the laser and material, often resulting in defects such as pore formation. Emissions collected during the process offer valuable insight but are difficult to interpret directly for defect detection. In this study, we propose a data-driven framework to interpret electromagnetic emissions in LBW using both supervised and unsupervised learning. Our framework is implemented in the post-process monitoring stage and can be used as a real-time framework. The supervised approach uses labeled data corresponding to predefined defects (in this work, pore formation is an example of a defined defect). Meanwhile, the unsupervised method is used to identify anomalies without using predefined labels. Supervised and unsupervised learning aims to find reference values in the emissions data to determine the values of signals that lead to defects in welding (enabling quantitative monitoring). A total of 81 welding experiments were conducted, recording real-time emission data across 42 spectral channels. From these signals, statistical, temporal, and shape-based features were extracted, and dimensionality was reduced using Principal Component Analysis (PCA). The LSTM model achieved an average mean squared error (MSE) of 0.0029 and mean absolute error (MAE) of 0.0288 on the testing set across five folds. The Isolation Forest achieved 80% accuracy and 85.7% precision in detecting anomalous welds on a subset with validated defect labels. The proposed framework enhances the interpretability of 4D photonic data and enables both post-process analysis and potential real-time monitoring. It provides a scalable, data-driven approach to weld quality assessment for industrial applications.

To advance quality assurance in the welding process, this study presents a deep learning (DL) model that enables the prediction of two critical welds’ key performance characteristics (KPCs): welding depth and average pore volume. In the proposed approach, a wide range of laser welding key input characteristics (KICs) is utilized, including welding beam geometries, welding feed rates, path repetitions for weld beam geometries, and bright light weld ratios for all paths, all of which were obtained from hairpin welding experiments. Two DL networks are employed with multiple hidden dense layers and linear activation functions to investigate the capabilities of deep neural networks in capturing the complex nonlinear relationships between the welding input and output variables (KPCs and KICs). Applying DL networks to the small numerical experimental hairpin welding dataset has shown promising results, achieving mean absolute error values of 0.1079 for predicting welding depth and 0.0641 for average pore volume. This, in turn, promises significant advantages in controlling welding outcomes, moving beyond the current trend of relying only on defect classification in weld monitoring to capture the correlation between the weld parameters and weld geometries.

The keyhole depth is a key measurement characteristic in the laser welding of busbar to battery tabs in battery packs for electric vehicles (EV), as it directly affects the quality of the weld. In this work, experiments are carried out with controlled and adjusted laser power and feed rate parameters to investigate the influence on the keyhole width, keyhole depth and porosities. A 3D numerical model of laser keyhole welding of an aluminum alloy (A1050) has been developed to describe the porosity formation and the keyhole depth variation. A new integration model of the recoil pressure and the rate of evaporation model is implemented which is closer to the natural phenomena as compared to the conventional methods. Additionally, major physical forces are employed including plume formation, upward vapor pressure and multiple reflection in the keyhole. The results show that keyhole depth is lower at higher feed rate, while lower feed rates result in increased keyhole depth. This study reveals that low energy densities result in an unstable keyhole with high spattering, exacerbated by increased laser power. Mitigating incomplete fusion is achieved by elevating laser energy density. The findings emphasize the critical role of keyhole depth in optimizing laser welding processes for applications like busbar-to-battery tab welding.

Today digital qualification tools are part of many design processes that make them dependent on long and expensive simulations, leading to limited ability in exploring design alternatives. Conventional surrogate modelling techniques depend on the parametric models and come short in addressing radical design changes. Existing data-driven models lack the ability in dealing with the geometrical complexities. Thus, to address the resulting long development lead time problem in the product development processes and to enable parameter-independent surrogate modelling, this paper proposes a method to use images as input for design evaluation. Using a case study on the curtain airbag design process, a database consisting of 60,000 configurations has been created and labelled using a method based on dynamic relaxation instead of finite element methods. The database is made available online for research benchmark purposes. A convolutional neural network with multiple layers is employed to map the input images to the simulation output. It was concluded that the showcased data-driven method could reduce digital testing and qualification time significantly and contribute to real-time analysis in product development. Designers can utilise images of geometrical information to build real-time prediction models with acceptable accuracy in the early conceptual phases for design space exploration purposes.

Many high-level technical products are associated with changing requirements, drastic design changes, lack of design information, and uncertainties in input variables which makes their design process iterative and simulation-driven. Regression models have been proven to be useful tools during design, altering the resource-intensive finite element simulation models. However, building regression models from computer-aided design (CAD) parameters is associated with challenges such as dealing with too many parameters and their low or coupled impact on studied outputs which ultimately requires a large training dataset. As a solution, extraction of hidden features from CAD is presented on the application of volume simulation of curtain airbags concerning geometric changes in design loops. After creating a prototype that covers all aspects of a real curtain airbag, its CAD parameters have been analyzed to find out the correlation between design parameters and volume as output. Next, using the design of the experiment latin hypercube sampling method, 100 design samples are generated and the corresponding volume for each design sample was assessed. It was shown that selected CAD parameters are not highly correlated with the volume which consequently lowers the accuracy of prediction models. Various geometric entities, such as the medial axis, are used to extract several hidden features (referred to as sleeping parameters). The correlation of the new features and their performance and precision through two regression analyses are studied. The result shows that choosing sleeping parameters as input reduces dimensionality and the need to use advanced regression algorithms, allowing designers to have more accurate predictions (in this case approximately 95%) with a reasonable number of samples. Furthermore, it was concluded that using sleeping parameters in regressionbased tools creates real-time prediction ability in the early development stage of the design process which could contribute to lower development lead time by eliminating design iterations. 

In this paper we present a semi-multiscale methodology, where a micrograph is split into multiple independent numerical model subdomains. The purpose of this approach is to enable a controlled reduction in model fidelity at the microscale, while providing more detailed material data for component level- or more advanced finite element models. The effective anisotropic elastic properties of each subdomain are computed using periodic boundary conditions, and are subsequently mapped back to a reduced mesh of the original micrograph. Alternatively, effective isotropic properties are generated using a semi-analytical method, based on averaged Hashin–Shtrikman bounds with fractions determined via pixel summation. The chosen discretization strategy (pixelwise or partially smoothed) is shown to introduce an uncertainty in effective properties lower than 2% for the edge-case of a finite plate containing a circular hole. The methodology is applied to a aluminium alloy micrograph. It is shown that the number of elements in the aluminium model can be reduced by 99.89 % while not deviating from the reference model effective material properties by more than 0.65 % , while also retaining some of the characteristics of the stress-field. The computational time of the semi-analytical method is shown to be several orders of magnitude lower than the numerical one. © 2021, The Author(s).

Neonatal sepsis is common, lethal, and hard to diagnose. In combination with clinical findings and blood culture, biomarkers are crucial to make the correct diagnose. A Swedish national inquiry indicated that neonatologists were not quite satisfied with the available biomarkers. We assessed the kinetics of 15 biomarkers simultaneously: ferritin, fibrinogen, granulocyte colony-stimulating factor (G-CSF), interferon (IFN)-γ, interleukin (IL)-1β, −6, −8, −10, macrophage inflammatory protein (MIP)-1β, procalcitonin, resistin, serum amyloid A (SAA), tumor necrosis factor (TNF)-α, tissue plasminogen activator-3 and visfatin. The goal was to observe how quickly they rise in response to infection, and for how long they remain elevated. From a neonatal intensive care unit, newborns ≥28 weeks gestational age were recruited. Sixty-eight newborns were recruited to the study group (SG), and fifty-one to the control group (CG). The study group subjects were divided into three subgroups depending on clinical findings: confirmed sepsis (CSG), suspected sepsis (SSG) and no sepsis. CSG and SSG were also merged into an entire sepsis group (ESG) for sub-analysis. Blood samples were collected at three time-points; 0 h, 12–24 h and 48–72 h, in order to mimic a “clinical setting”. At 0 h, visfatin was elevated in SSG compared to CG; G-CSF, IFN-γ, IL-1β, −8 and − 10 were elevated in SSG and ESG compared to CG, whereas IL-6 and SAA were elevated in all groups compared to CG. At 12–24 h, IL-8 was elevated in ESG compared to CG, visfatin was elevated in ESG and SSG compared to CG, and SAA was elevated in all three groups compared to CG. At 48–72 h, fibrinogen was elevated in ESG compared to CG, IFN-γ and IL-1β were elevated in SSG and ESG compared to CG, whereas IL-8 and SAA were elevated in all three groups compared to CG. A function of time-formula is introduced as a tool for theoretical prediction of biomarker levels at any time-point. We conclude that SAA has the most favorable kinetics regarding diagnosing neonatal sepsis, of the biomarkers studied. It is also readily available methodologically, making it a prime candidate for clinical use.