Machine learning applied for process design of hybrid metal-polymer joints

Abstract

Several studies demonstrated a close relationship between the strength of metal-polymer joints made by thermo-mechanical processes and the temperature reached during the process. Machine Learning techniques have been adopted to forecast the temperature trends produced during Friction Assisted Joining varying the main process parameters. This was performed in four steps. First, the design of a campaign of experimental tests and the setup of an instrumented joining machine equipped with load sensors (plunging load and torque) was implemented. The machine was also equipped with an IR camera for temperature measurements during the process. Then, the experimental data were processed and filtered to produce a robust dataset, which was subsequently used for the training, calibration, and validation of an Artificial Neural Network (ANN). The developed approach was based on a data encoding-decoding approach that initially subdivided power-temperature data for training the network. During training, the ANN had no information on the origin curve. Then, once validated, the ANN reconstructed the temperature curve sequentially starting from power curve measurements. The results indicated that starting from the absorbed (measured) power, the developed model was capable to predict the temperature evolution with great accuracy. The resulting Artificial Neural Network represents a reliable tool for process design and identification of correct process parameters e.g. Dwell time; besides, it can be used during the joining process to achieve a prescribed temperature.