Deep Learning is a powerful tool for industry and other sectors that benefits from large datasets and computing capacity during the design and training steps. CESGA’s FinisTerrae HPC infrastructure can reduce enormously the time to solution thanks to its parallel capabilities and the availability of fast and large storage. As a demonstration of these benefits, CESGA has published a white paper (as a technical report) with a successful industrial case of the benefits of High Performance Computing on Machine and Deep Learning applications. The report is based on an use case from the experiment 707: “Cyber-Physical Laser Metal Deposition (CyPLAM)” developed under the framework of the Fortissimo 2 project (H2020). This experiment has demonstrated the possibilities of this technology for complex high-tech industrial technologies control and monitoring. The Machine Learning model has been developed by AIMEN Technology Centre and tested at EMO manufacturing plant in Slovenia.

In this experiment, a hyper-parametric search the best Convolutional Neural Network model, for a monitoring system of Laser Metal Deposition (LMD) based on Medium Wavelength Infrared (MWIR) images, was conducted on the FinisTerrae II (FT2) infrastructure using the Google’s Machine Learning API Tensorflow. Thanks to the use of HPC this search could be reduce from 40 days of computation to less than 70 hours.