A Coupled Approach Based on Statistical Methods and Machine Learning Techniques to Improve Porthole Die Design

Bulk metal forming techniques provide advantageous production routes in the modern manufacturing industry. Although the scientific literature has offered continues advances in the process design, the well-known bulk forming methods, including extrusion, forging, drawing and rolling, still require update and optimisation. In this study, the attention was pointed out on porthole die extrusion employed for production of cross-sectional hollow profiles. Coupled statistical and machine learning techniques were implemented with the aim to predict the output of the process and to allow a performant design of the porthole die before its manufacturing. Specifically, a dataset was considered identifying 12 geometric variables, each one varied on three levels, starting by a specific case study, where a standard porthole die is employed to extrude circular section profiles. Then, employing the process data extracted by literature, a statistical analysis was preliminary carried out to highlight the relationships among the most significant variables that can affect the outcome modelling. Subsequently, a set of models based on machine learning techniques was trained and tested to well generalize the behaviour of the process, changing the investigated factors. The considered methodologies were implemented on R-Studio software. The aim of the study is the rapid control of any alteration of the standard processing conditions, giving flexibility to the work, ensuring the correct design of the porthole die in a shorter time and saving money. All the details will be provided in the manuscript.