Abstract
Utilizing the high potential of lightweight components consisting of similar or dissimilar materials can be achieved through solid-state joining techniques. As opposed to fusion-based joining processes, typical defects such as pores or hot-cracking can be prevented and high quality welds are obtained. However, determining the optimal process window for desired joint properties is a time consuming task and requires many experiments.
Data-driven machine learning models exhibit strong capabilities to identify and quantify relationships along the process-structure-property-performance chain. Integrating corresponding microstructural characteristics through classification and segmentation of salient microstructural features can help further improve the precision of those relationships. Predictions of mechanical properties based on particular process parameters but also inverse determination of required process parameters for desired properties can be accomplished. The generated knowledge can deepen the understanding of underlying mechanisms, identifying optimized parameter sets for new material combinations, while simultaneously reducing the demand for a fast number of experiments to be conducted.
Presented will be results from the application of various machine-learning-models to correlate required process parameters with desired joint properties for two solid-state joining techniques: Refill Friction Stir Spot Welding and Friction Riveting. The employed machine-learning-models for regression and classification tasks include decision trees, random forests and support vector machines as well as convolutional neural networks. Training and testing data was generated through Central-Composite and Box-Behnken Designs of Experiments. The models were trained and tested based on different performance measures in order to evaluate the suitability of the different approaches for the current processes. The results illustrate the predicting capabilities with respect to certain process parameters on mechanical properties and performances, which enable inverse identification of optimized process parameters to manufacture desired mechanical properties.
