Data-Driven Estimation of Droplet Formation for Conductance Control in Wire-Based Metal Additive Manufacturing

Abstract

This work studies a wire-based metal additive manufacturing process where the distance between the wire and the part being built is controlled indirectly through electrical conductance. The measured conductance is compared to a setpoint, and the wire is raised or lowered depending on whether the conductance is too high or too low. At present, this setpoint is adjusted manually by an operator monitoring the process. The aim of this work is to investigate whether the conductance setpoint can instead be determined automatically, using models that estimate the probability of droplet formation from measured process data. Several approaches are evaluated, including linear models based on PCA as well as convolutional neural networks. The linear models show some correlation with droplet formation but are generally noisy. The convolutional neural network models show promising preliminary results, although further work is required before it can be used as the basis for a control algorithm. In addition, a simple control algorithm based only on observed drops, rather than predictions of future drops, was evaluated. The algorithm was tested in only one experiment, so the results should be considered preliminary. Nevertheless, the results indicate that the algorithm was able to control the process with a similar performance to that of a human operator.