On the maintenance of processing stability and consistency in laser-directed energy deposition via machine learning

In lateral wire-based laser-directed energy deposition, conveying shielding gas through the wire feed nozzle devastates the processing stability, which results in a geometrical deviation and an increase in porosity level. In the present study, an extra nozzle is installed to convey shielding gas to balance the gas flow from the wire feed nozzle. It is confirmed that the installation of the extra nozzle sustains the processing stability, achieves geometrical accuracy, and reduces the porosity level. However, finding an appropriate flow rate for the extra shielding gas is time- and material-consuming. In order to efficiently find the flow rate, a convolutional neural network is used to simplify this process by analyzing the processing images and receiving guidance from the outputs to adjust the current flow rate to save time and material cost. In addition, a novel methodology is proposed to in-situ monitor and in-situ adjust process parameters during laser-directed energy deposition by adopting a convolutional neural network. The processing characteristics such as melt pools, plume, and spatter can be well maintained, which contributes to a consistent geometry and porosity of deposition layers. Results indicate that the methodology proposed in this study is promising to be transferred to other laser-beam-melting processes both in additive manufacturing and coating.