Weld Bead Size Modeling And Parameter Prediction For Gmaw-based Rapid Manufacturing Through A Neural Network

The GMAW rapid prototyping is a new processing method based the discreteaccumulation principle. To the background of the intelligent rapid prototypingprocess, in order to predict the cladding process parameters, this paper employs theneural network modeling method to discuss the relationships between the claddingprocess parameters and the bead geometry in the rapid prototyping to realize thefunction of calculating the bead geometry according to the given cladding processparameters and reasonably choosing the cladding process parameters on thecondition of needed shaping sizes.First of all, this paper uses the wire feed speed, welding speed, the voltage andthe distance of nozzle to plate as the input variables, at the same time, the width andheight of the weld bead are used to be the output variables. The test sample isdesigned with the method of quadratic regression general rotation design and thesingle bead experiment is taken on the substrate. The width and height of everybead are measured with the structured light vision sensor and the complete data canbe obtained.The regression equation between the cladding bead width, height and theinput variables is established with the traditional quadratic regression method,respectively. The statistical test is carried out and the insignificant item is removedso as to get the optimization model considering the width and height and the inputvariables.Taking into account the highly nonlinear between the output variables and theinput variables, the nonlinear forward model of the multi-input multi-outputvariables is constructed with the Artificial Neural Networks. The neural networkmodel has the higher accuracy compared to the traditional quadratic regressionmethod with the designed verification samples. After that, the simulation of thecladding process behavior is conducted on the basis of the established neuralnetwork model and the effects of the cladding process parameters on the bead sizesare revealed.In the actual prototyping process, the cladding process parameters need bepredicted by the ideal bead sizes. Therefore, the inverse model of the neuralnetwork is established with the bead sizes as the input variables and the processparameters as the output variables. As a result of the restriction of the inverse model, the closed-loop feedback iterative system combining the forward model and theinverse model is established. This feedback system has been proved to be reliablethrough test and is able to applied to predict the cladding process parameters.To further verify the reliability of the combined system and the significance ofthe first weld bead geometry to the entire geometry of the bead, the single-channelmulti-layer experiment is carried out on the thin-walled flatbed. Through comparingthe deviation between the fires bead and the forming size eventually, the correctnessand necessity of prediction of the process parameters is proved.