Self-adjustable tool center point of industrial robots based on neural network models

The work addresses the problem of filler wire deviation from the tool center point in Wire Arc Additive Manufacturing based on Gas Metal Arc Welding. The primary causes identified for this deviation are inherent wire curvature and the progressive wear of the welding torch's contact tip, leading to an offset between the actual material deposition location and the programmed trajectory. This study analyzes and identifies a series of parameters that most significantly influence the magnitude of the filler wire deviation. It was determined that the wire-tip contact interaction is characterized by the wire's stress-strain state, which governs the dynamic changes in contact force throughout the tip wear process. To establish the relationship between the deviation and the key parameters of this contact interaction, effective flexural modulus, microhardness, diameter, and radius of curvature were determined for Inconel 718, Inconel 625, and low-carbon steel G3Si1 welding wires. Based on measured properties, a simulation of the wire-tip contact interaction was developed to determine contact forces, subsequently validated by laboratory experiments. Using the experimental data set and applying artificial intelligence methods, a system of three multi-layer perceptron neural network models was developed. The use of neural networks enabled the description of the non-linear relationship between the selected parameters and allowed for the prediction of wire deviation with an experimentally measured mean absolute error of 0.088 mm. The resulting models generalize and comprehensively describe the dependency of the wire deviation magnitude relative to the tool center point, utilizing a minimally necessary set of input parameters.