Improving Metal Stamping Efficiency via Neural Networks

A team of University of Guelph researchers are aiming to make manufacturing processes more efficient. Hot stamping, also known as press hardening, is a globally adopted manufacturing process used to produce ultra-high strength steel parts for the automotive industry. To prevent oxidation — the chemical process behind corrosion — during the heating process, hot stamping steels are Aluminium-Silicone (Al-Si) coated. Prior to hot stamping, the sheet metal blanks go through a heating process to approximately 900°C in which the thin Al-Si coating melts and reacts with the base metal to form complex Fe-Al and Fe-Al-Si intermetallic compounds. The heating process is then followed by a soaking stage until the blank reaches a target temperature.

The furnace (gas-fired) time used for heating is an expensive process but necessary to ensure the coating transforms into intermetallics and the base metal austenitizes (a process where the metal’s crystal structure is changed). A current focus in hot stamping research attempts to predict the different intermetallic phases formed during the heating and soaking cycle. The intermetallic coating affects the hot stamped parts’ weldability, paintability, and corrosion protection to some degree. A professor and researcher at the University of Guelph have instead chosen to take a new route to predict intermetallic coating growth.

Testing the Heating and Soaking Process

University of Guelph School of Engineering professor Dr. Alexander Bardelcik, University of Guelph School of Computer Science professor Dr. Neil Bruce and PhD student Siyu Wu collaborated with Constantin Chiriac of Ford Motor Company and Cangji Shi of Magna International to develop a new coating growth model for improving the hot stamping process. This research investigates the effects of furnace heating rates (1-7.4°C/s econd) for target blank temperatures that range from 600-1000°C. Through the heating rate investigations, Bardelcik and Wu were able to identify and quantify the complex growth of different intermetallic phases (using elemental composition) throughout the coating thickness and use the results to train, test, and validate a neural network model. 

In addition to testing the heating rates, Bardelcik and Wu assessed the effect of soaking time and temperatures on the intermetallic growth. The tests consider soaking times ranging between 30-240 seconds and soaking temperatures from 900-1000°C. The effect of soaking time on the intermetallic growth of the coating was also used to train, test, and validate the neural network model.

Creating Consumer Profiles Using Machine Learning

The experimental results found in the heating and soaking testing phases were used to develop a deep neural network (DNN) to  predict the coating chemical composition with respect to the heating rate and soaking temperature / time. A deep neural network is a subset of neural networks with a more complex structure than traditional neural networks. The DNN accepts four parameters: the heating rate, heating time, soaking time and the distance of each elemental weight composition (Al, Si, Fe) from the coating surface. There were two models developed for each coating weight. The model was able to predict the proper intermetallic phase  transformations for three simulated heating rates (1.5, 2.4 and 7.3°C/s). Testing these heating rate simulations showed that the furnace time can be reduced, and a good coating can still be achieved.

“This model will reduce furnace cycle time without having to do trial and error,” says Bardelcik. “It helps with making the profile of the coating layers optimised for better welding, painting, and to prevent corrosion”.  

Bardelcik also mentions that their model predicts the diffusion layer thickness, a layer between the coating and the base metal. Since the coating layer will always crack, a thicker diffusion layer will better protect the base metal. Their work has significant applications in the automotive, rail, and agricultural machinery industry. 

This story was written by Izabela Savić as part of the Science Communicators: Research @ CEPS initiative. Izabela is a PhD student at the School of Computer Science under Dr. Daniel Gillis. Her research focus is on cybercrime and better creating tools to better equip law enforcement and user security.

The work on distinguishability in quantum communication is supported by the National Sciences and Engineering Research Council (NSERC) of Canada.

Reference: S. Wu, Z. Zhou, N. Bruce, A. Bardelcik, C. Chiriac, C. Shi, “A simulation of Al-Si coating growth under various hot stamping austenitization parameters: An artificial neural network model,” Materials Today Commnucations, vol. 38, 2024. Doi: https://doi.org/10.1016/j.mtcomm.2024.108492.