Towards neural network-augmented adaptive control of deposition height stability in alternating current arc-directed energy deposition of aluminum alloys based on in-situ arc voltage sensing

Alternating current (AC) gas tungsten arc-directed energy deposition (GTA-DED) demonstrates broad application prospects for fabricating aluminum alloy parts. However, the high thermal conductivity and low liquid viscosity of aluminum alloys deteriorate molten pool stability and pose significant challenges to accurate forming of aluminum alloys in GTA-DED. Targeting the challenges of signal noise and forming quality in AC GTA-DED of aluminum alloys, a novel recursive filtering strategy and a neural network-augmented adaptive controller are developed to achieve precise processing of variable-polarity arc voltage (AV) signals and accurate control of deposition height. In this study, the positive peak AV is extracted from the raw AV signals, and a Kalman-alpha filter is designed to remove the noises contained in AV signals, obtaining a high signal-to-noise ratio (SNR) of 27.68. A linear model between AV and arc length is established to characterize the deposition height stability. The system's dynamic characteristics are analyzed via step experiments with wire feed speed (WFS) as the input and positive peak AV as the output. A control system based on the backpropagation neural network Proportional-Integral-Derivative (BPNN-PID) controller is designed to real-time regulate WFS to improve the deposition height stability, which helps advance the application of artificial intelligence in GTA-DED. After applying the control system, the height deviation of the 76-layer thin-walled part is less than 0.46 mm (mm), illustrating excellent control effects. This study is expected to provide valuable insights and guidelines for the intelligent control of GTA-DED of aluminum alloys.