Laser welding is a manufacturing process widely used in the industry to efficiently join parts together, generally through a characteristic deep penetration melt pool. Its benefits, like no-contact welding, no tool wear and fast processing, are very relevant for many industrial applications. Nonetheless, finding the optimal parameters for each specific processes remains challenging and time-consuming. Involving many physical phenomena, such as laser-matter interaction, thermodynamics and fluid mechanics, the process parameters have many nonlinear interactions. In these circumstances, a cost and time-effective Design of Experiment (DoE) is nearly impossible to generate. Furthermore, thorough weld characterisation, from geometrical to metallurgical analysis, remains a labor-intensive and expensive task. In this study, we compared different regressors powered by Artificial Intelligence such as Gradient Boosted Decision Trees, Gaussian Process Regressors, Perceptrons trained on readily available data from previous trials done at IREPA LASER, to predict the depth of penetration of the weld. To develop the model with industrial use in mind, the material, the processing parameters and the optical setup were used as the input parameters. A R2 of 0.94 and a Mean Squared Error of 0.25mm2 are obtained from the model developed. Scores are then compared to the state of the art, taking into consideration the size and number of parameters of the dataset used.
|