Although applications of additive manufacturing (AM) have been significantly increased in recent years, its broad application in several industries is still under progress. AM also known as three-dimensional (3D) printing is layer by layer manufacturing process which can be used for fabrication of geometrically complex customized functional end-use products. Since AM processing parameters have significant effects on the performance of the printed parts, it is necessary to tune these parameters which is a difficult task. Today, different artificial intelligence techniques have been utilized to optimize AM parameters and predict mechanical behavior of 3D-printed components. In the present study, applications of machine learning (ML) in prediction of structural performance and fracture of additively manufactured components has been presented. This study first outlines an overview of ML and then summarizes its applications in AM. The main part of this review, focuses on applications of ML in prediction of mechanical behavior and fracture of 3D-printed parts. To this aim, previous research works which investigated application of ML in characterization of polymeric and metallic 3D-printed parts have been reviewed and discussed. Moreover, the review and analysis indicate limitations, challenges, and perspectives for industrial applications of ML in the field of AM. Considering advantages of ML increase in applications of ML in optimization of 3D printing parameters, prediction of mechanical performance, and evaluation of 3D-printed products is expected.

Machine learning in predicting mechanical behavior of additively manufactured parts

Additive manufacturing (AM) is poised to bring a revolution due to its unique production paradigm. It offers the prospect of mass customization, flexible production, on-demand and decentralized manufacturing. However, a number of challenges stem from not only the complexity of manufacturing systems but the demand for increasingly complex and high-quality products, in terms of design principles, standardization and quality control. These challenges build up barriers to the widespread adoption of AM in the industry and the in-depth research of AM in academia. To tackle the challenges, machine learning (ML) technologies rise to play a critical role as they are able to provide effective ways to quality control, process optimization, modelling of complex systems, and energy management. Hence, this paper employs a systematic literature review method as it is a defined and methodical way of identifying, assessing, and analysing published literature. Then, a keyword co-occurrence and cluster analysis are employed for analysing relevant literature. Several aspects of AM, including Design for AM (DfAM), material analytics, in situ monitoring and defect detection, property prediction and sustainability, have been clustered and summarized to present state-of-the-art research in the scope of ML for AM. Finally, the challenges and opportunities of ML for AM are uncovered and discussed. 

Research and Application of Machine Learning for Additive Manufacturing

Machine learning for metal additive manufacturing: Towards a physics-informed data-driven paradigm

http://manuscript.elsevier.com/S1526612522000093/pdf/S1526612522000093.pdf

Machine learning algorithms for defect detection in metal laser-based additive manufacturing: A review

Additive manufacturing (AM) is gaining popularity because of its ability to manufacture complex parts in less time. Despite recent research involving designs of experiments (DOEs) to characterize the relationships between some AM process parameters and various part quality characteristics, to date, there seems to be no universally accepted comprehensive model that relates process parameters to part quality. In this paper, to support the goal of manufacturing parts right the first time, a Bayesian network in continuous domain is developed which relates four process parameters (laser power, scan speed, hatch spacing, and layer thickness) and five part quality characteristics (density, hardness, top layer surface roughness, ultimate tensile strength in the build direction and ultimate tensile strength perpendicular to the build direction). A machine learning algorithm is used to train the network on a database mined from a large number of publications with experimental data from parts built using 316L with selective laser melting. Using this Bayesian network, the user is able to enter a value for one or more known nodes or variables, and the network provides predictions on all the remaining nodes in the form of probability distributions. A method is developed whereby the user inputs are checked for reasonableness using an n -dimensional convex hull, and if necessary a recommendation is returned based on user-defined weights. The network is validated by retaining a subset of the training data for testing and comparing the network’s predictions to the known values. Accuracy is optimized by continually re-training the network using parts built with a specific machine of interest. The industrial relevance of this research is outlined with respect to four current challenges in AM, including the length of time to determine optimal process parameters for a new machine, ability to organize relevant knowledge, quantification of machine variability, and transfer of knowledge to new operators.

Prediction of selective laser melting part quality using hybrid Bayesian network

Generative adversarial network for early-stage design flexibility in topology optimization for additive manufacturing

Multi-Material Topology Optimization Using Variable Density Lattice Structures for Additive Manufacturing


 Additive manufacturing has enabled the production of intricate lattice structures that meet stringent design requirements with minimal mass. While many methods such as lattice-based topology optimization are being developed to design lightweight structures for static loading, there is a need for design tools for achieving dynamic loading requirements. Lattice structures have shown particular promise as low-mass energy absorbers, but the computational expense of nonlinear finite element analysis and the difficulty of obtaining objective gradient information has made their optimization for impact loading particularly challenging. This study proposes a Bayesian optimization framework to determine the lattice structure design that provides the best performance under a specified impact, while managing the structure’s mass. Considering nonlinear effects such as plasticity and strain rate sensitivity, a 2D explicit finite element (FE) model is constructed for two lattice unit cell types under impact, and parameterized with respect to geometric attributes such as height, width, and strut thickness. These parameters are considered design variables in a minimization problem with an objective function that balances part volume with a common injury metric, the head injury criterion (HIC). Penalty values are assigned to designs that fail to absorb the entire impact. Design for additive manufacturing (DFAM) constraints including minimum feature thickness and maximum overhang angle are applied to ensure that the optimal design can be manufactured without subsequent manual refinement or post-processing. The best optimizer hyperparameters are then carried over into larger optimization problems involving lattice structures. Future work could include expanding this framework to allow for lattice structure designs with arbitrary boundaries.

Bayesian Optimization of Energy-Absorbing Lattice Structures for Additive Manufacturing


 Material penalization and filtering schemes are key strategies applied to topology optimization (TO) to promote more discrete and manufacturable designs. However, these modifications introduce fluctuations in the design landscape that amplify non-convexity and influence the local minima identified by TO. Harnessing the machine learning approach of generative adversarial networks (GAN), we investigate the role of penalization and filtering by comparing the designs between TO and GAN-based TO surrogates. A total of 17 GANs were constructed to predict 2D minimum compliance topologies across a set of penalization factors and filters, each interpolating a design space of 270,000 boundary condition and loading scenarios. The prevalence of GAN-predicted topologies with better compliance than TO-calculated topologies was estimated via a random sampling of the design space. GAN ‘over-performance’ occurs across material penalization and filtering conditions, where the frequency tends to increase as penalization increases. Analysis of this test set is leveraged to highlight trends regarding the conditions under which this ‘over-performance’ occurs, and the geometric characteristics these designs exhibit. Collectively, this study presents an alternative method to characterize the effects of penalization and filtering on design outcomes and motivates the use of data-driven surrogates to augment traditional approaches.

Nathan Hertlein

Papers

Prediction of selective laser melting part quality using hybrid Bayesian network

Generative adversarial network for early-stage design flexibility in topology optimization for additive manufacturing

Multi-Material Topology Optimization Using Variable Density Lattice Structures for Additive Manufacturing

Bayesian Optimization of Energy-Absorbing Lattice Structures for Additive Manufacturing

Generative Adversarial Design Analysis of Non-Convexity in Topology Optimization