In a recent article published in the journal Additive manufacturingresearchers discussed machine learning prediction of melt pool characteristics (MeltpoolNet) in metal additive manufacturing.
Study: MeltpoolNet: predicting melt pool characteristics in metal additive manufacturing using machine learning. Image Credit: Nordroden/Shutterstock.com
Metal Additive Manufacturing (MAM), a new and innovative manufacturing technology that enables the manufacture of sophisticated components, has paved the way for the next industrial revolution. MAM can manufacture parts and components with complex shapes using special metals and materials.
Increasing the scale, speed and quality of printed products remains a huge problem, despite the fact that additive manufacturing has attracted a lot of scientific and technical work in industry and academia. Various experimental tracking approaches are frequently used to find the optimal processing window for manufacturing low-defect parts.
Due to the multi-physics and multi-scale nature of AM processes, as well as the major impact of processing parameters on printed products, AM has evolved from purely physics-derived approaches to ones based on data and physics. As a result, machine learning (ML) and data-driven analysis have become standard procedures in advanced manufacturing applications, and AM research is becoming increasingly popular. Creating ML models for metal additive manufacturing, on the other hand, requires overcoming the hurdles of data scarcity.
About the study
In this study, the authors discussed a comprehensive methodology for benchmarking ML for molten pool characterization. More than 80 MAM publications have been used to compile an extensive experimental data set including MAM processing parameters, weld puddle size, materials, weld puddle modes and defect types. To establish a comprehensive learning framework for defect and weld pool shape prediction, physics-aware MAM characterization, evaluation measurements and adaptable ML models were used.
The benchmark used by the authors was based on a larger data set, which allowed for the improvement and regulation of component quality over a wider range of materials and processing parameters. The proposed dataset contained a minimum of 80 sources of experimental data gathered from studies performed on a small number of alloys using a single AM method. These various sources of data have been combined into a larger repository for the characterization of weld pools across a diverse range of alloys and processing parameters. A set of AM machine learning methods called MeltpoolNet has been created to characterize the behavior of the melt pool.
The researchers used MeltpoolNet to create ML models using the full experimental dataset. MeltpoolNet enabled the optimization of melt pool processing parameters and the prediction of porosity in printed products. Experiments with various materials, processing parameters and MAM process types were included in the dataset.
The impact of various construction process parameters on the performance of the proposed models was also illustrated. In addition, a data and model-based identification method was created to discover clear relationships between dataset processing parameters and material qualities, which was more interpretable than ML models.
Classification job accuracy was 85.6% when the XGBoost model was used to predict weld pool classification using baseline parameters, absorption coefficients, and elemental characterization. With a baseline feature model configuration and hardware hot coding as input to the random forest algorithm, the best result was achieved with an accuracy of 85.78%. The gradient amplification model achieved an accuracy of 99.55% and a mean absolute error (MAE) of 10.92 µm.
Gradient boosting, neural networks, and random forest have defeated other ML models for regression and classification. Feature engineering in AM has also been shown to be essential for producing a generalizable and highly accurate ML model.
The proposed explicit models were not only more interpretable than the used ML models, but also had better weld pool geometry prediction ability than Rosenthal’s weld pool geometry estimation.
In conclusion, this study presented a comprehensive machine learning benchmark for predicting weld pool geometry and defect type. Data from a wide range of AM melt pool characterization experiments have been collected. To improve the accuracy of different ML models, various feature engineering approaches for AM input data have been introduced. Several machine learning models were compared to each other using different characterization strategies. In addition, evaluation measures and reporting standards were also explored.
In addition, explicit data-driven models for estimating weld pool geometry from process parameters and material properties were found, which exceeded Rosenthal’s estimate for weld pool geometry while maintaining interpretability. Additionally, based on the dataset processing parameters and material attributes, a data-driven pattern identification method was created to estimate the weld pool geometry.
The authors believe that this repository can be used to control weld pools and optimize processes. They expect that by providing a consistent platform for comparison and evaluation, this benchmark will facilitate the optimization and regulation of additive manufacturing processes and that MeltpoolNet will become a comprehensive resource for the machine learning community. metal additive manufacturing.
More from AZoM: How is the fast Fourier transform used alongside electron microscopy?
Akbari, P., Ogoke, P., Kao, NY., et al. MeltpoolNet: predicting melt pool characteristics in metal additive manufacturing using machine learning. Additive Manufacturing 102817 (2022). https://www.sciencedirect.com/science/article/pii/S2214860422002172