On Monday the 16th Sept at 2pm in G.03 there will be a CISA Seminar by Tudor George Alexandru who will be speaking on 'An approach for facilitating thermal design through the use of intelligent systems for Industry 4.0'.
Industry 4.0 aims to develop interconnected manufacturing systems through the ongoing improvement of products and product related services. From an enterprise perspective, this industrial transformation requires the widespread use of real time networking supported by intelligent systems. Large amounts of workshop floor data need to be collected and then used for training models which can help predict failure and produce automatic feedback. To make Industry 4.0 possible, intelligent systems must be successfully integrated into manufacturing, which relies on the ability of the stakeholders to transform conventional workshop floor devices to cyber-physical ones.
One difficulty is represented by arising thermal issues, due to the tight space and power constraints applied, combined with environmental policies and unpredictable device obsolescence. In particular, thermal design has become a limiting aspect of the cyber-physical system development workflows. From an engineering point of view, thermal design can be decided on using complex multiphysics simulations. However, the required knowledge and resources needed for such simulations are only available to high profile companies. This research proposes a machine learning based approach that combines experimental knowledge with a lumped parameter heat transfer simulation model to reduce interdisciplinary variables to tangible quantities, thus allowing for fast and accurate thermal design solutions to be achieved. To this end, a thermal characteristics dataset was first built for a batch of CPUs using known and experimental data. Machine learning regression was then deployed to predict !
temperature dependent variables. The results were used for parameterizing capacitance and conductance in a transient thermal simulation. This allowed for the total heat transfer rate to be estimated by knowing the load imposed by the embedded software on each core. From this stage, all the necessary data for the sizing process was derived, allowing for a precise cooling component selection to be performed.
Alexandru Tudor George is a 3rd-year PhD Student at the Politehnica University of Bucharest, Faculty of Industrial Engineering and Robotics, where he also received a BEng and MSc degree. His research focuses on the use of Computer Aided Engineering software for solving industry related problems by formalizing multiphysics viewports. He also works as a Technical Specialist in the field of Aerospace Stress Engineering for Expleo Romania, being involved in the development and optimization of commercial aircraft structures. He is a Student Member of the IEEE Computer Society and Member of the Association ICMAS.
Refreshments will be available in the room after the Seminar.