Prediction of melt pool shape in additive manufacturing based on machine learning methods.

• Building three optimized machine learning methods to predict melt pool shape. • Utilization of melt pool shapes derived from a large number of directed energy deposition experiments as data sets. • Prediction accuracy varies between models applying different algorithms, with accuracy of up to 96% or more. Directed Energy Deposition (DED), one type of additive manufacturing (AM) as novel modern manufacturing technology, is widely employed to fabricate materials layer by layer through digital models. It is well known that the shape of the melt pool is influenced by the manufacturing process parameters during the manufacturing process, which can affect the performance of the part. Therefore, three machine learning models, support vector regression (SVR), extreme gradient boosting (XGBoost), and back propagation neural network (BPNN), were selected and constructed in this study to predict melt pool shape and to maximize prediction performance by integrating various algorithms and parameters. Meanwhile, the geometry of the melt pool, such as the height, width, and depth of 210 single-channel deposition layers, were measured as training and testing datasets, which can be supplied the basic data for training the prediction models and testing the prediction performance of models. In addition to the training and test datasets, a new dataset with 36 samples was created to validate the performance predictions made by the three machine learning models. The results demonstrate that, based on the test dataset, the RBF kernel-based SVR model predicts the melt pool height with an accuracy of ∼ 93 %, while the XGboost model predicts the melt pool width and depth with accuracy rates of ∼ 97 % and ∼ 96.3 %, respectively. In the new dataset, the BPNN model based on Adam's method achieves a prediction accuracy of ∼ 93.7 % for melt pool height, but the XGboost model achieves ∼ 96.6 % and ∼ 97.8 % for melt pool width and depth, correspondingly. Thus, in addition to the training and testing datasets, the enhanced machine learning models exhibited remarkable prediction accuracy, excellent generalization, and robustness in the new dataset. As a result, the application of machine learning has greatly improved the possibility of controlling the DED process more intelligently and stably. [ABSTRACT FROM AUTHOR]