Inconel-718 Processing Windows by Directed Energy Deposition: a Framework Combining Computational Fluid Dynamics and Machine Learning Models with Experimental Validation

In directed energy deposition (DED), it is widely known that the printed layer dimensions and morphology are highly affected by the process parameters, which affect the performance of the in-service part. Various machine learning (ML) models have been presented for deposited layer dimensions; however, the determination of an optimum model is still lacking. The purpose of this work is to explore the widely applicable ML models, including Gaussian process surrogate (GPS), extreme boosting gradient (EBG), and support vector machine (SVM), to predict the deposited layer dimensions accurately without compromising the performance. In this study, a multi-physics CFD simulation model was applied using Flow-3D software for the Inconel-718 DED processing incorporating the volume of fluid and discrete element modeling techniques. After experimental validation of the proposed CFD model, the geometry of 343 single-layer depositions, including width and height, were measured for training, and testing of ML models. Following the ML model training and testing, processing windows have been developed for the deposited layer’s width and height, by varying operating conditions. The results demonstrated that the GPS model has the best overall prediction, with the R -squared value = 0.99 for the deposited layer dimensions. After GPS, SVM presented results with an R -squared value of 0.98 for deposition layer dimensions. EGB presented the least accurate results showing the R -squared value of 0.94 and 0.81 for layer width and height, respectively. This study provides a framework for the smartification of printed layer geometry estimation and control for DED by combining modeling, experiments, and machine learning.