There are currently 5 industrial success stories presented by NCC Turkey demonstrating the benefits of HPC-based tools and techniques in innovative SMEs as well as large industrial companies. Excellent results were obtained for the adoption of HPC in a variety of sectors and domains such as materials modelling, computational fluid dynamics, high-performance data analytics, and artificial intelligence.
NANOGRAFI: Improving the Efficiency of the Graphene-Enhanced Polymer Composite Production via Classical Molecular Dynamics
Nanografi is a company that specializes in the production of nanomaterials including graphene- and graphene oxide-enhanced polymer composites. The company’s protocol is based on a “synthesize-test-improve” with a lot of the human resources spent in the trial-and-error stages to develop the composite with the best mechanical properties. Since there are a large number of parameters that potentially go into this problem such as the type of the polymer, the sheet number of graphene, and pretreatment conditions, these experiments are both costly and time-consuming. Our team has proposed an alternative mechanism where the initial elimination of the composites to be synthesized is to be simulated using classical molecular dynamics. With this case study, the company was introduced to computational materials science and HPC services for the first time and was able to tap into computational work as an alternative or support mechanism to the traditional and rather time-consuming experimental methods. With the support of NCC Turkey, Nanografi applied for FF4EuroHPC Second call and the company’s application was accepted. The case study collaboration with METU and TRUBA pawed the way for the successful FF4EuroHPC application.
End-user: Nanografi
Domain expert: Middle East Technical University
HPC infrastructure provider: TUBITAK ULAKBIM TRUBA HPC Center
Success story details: http://eurocc.truba.gov.tr/?page_id= 6181&lang=en
Keywords: Manufacturing, materials science, HPC
MACHINETUTORS: Large Scale Real-Time Image Content Moderation
Founded in 2010, Machinetutors provides machine learning consultancy and customized AI software development services. This project addresses the problem of large-scale real-time image-based content moderation. The system is deployed to a production environment where tens of thousands of users browse the internet daily. The system must be both accurate and run in real-time to meet the business requirements. Three main models were developed. In the first model, a multi-label NSFW classifier that can detect the NSFW levels (light, medium, hard) and predict other labels, such as the real person and clothing characteristics were proposed. The second model was a one-stage body -based age & gender detection model. The third model was a segmentation model.
With this case study, the company was able to run many experiments in parallel and quickly see the effects of the model updates. Being able to access many GPUs at the same time enabled the company to tune the hyper-parameters of each model to improve the results. Besides the speed and cost-efficiency provided by this support have helped the company gain a considerable competitive advantage in the global AI ecosystem.
End-user: MachineTutors
HPC infrastructure provider: TUBITAK ULAKBIM TRUBA HPC Center
Success story details: http://eurocc.truba.gov.tr/?page_id=6297&lang=en
Keywords: Social media, computer vision, machine learning, deep learning, artificial intelligence
DSTECH: Simulation-Optimisation of a Patented Design with Parallel Computing on TRUBA cluster – Design-Optimisation of a Disinfection System
DSTECH is an SME company supplying services to customers in the solution of complex engineering problems encountered in different disciplines such as Environment, Energy, and Aerospace sciences. In this case study, DSTECH has developed an open-source simulation and optimisation framework augmented with OpenFoam, Dakota, and Python, which can be freely used on HPC clusters without any restrictions such as license, commercialization, and parallelization. The present simulation and optimisation framework was employed to optimise the slot-baffle design in three stages without any convergence issues. Numerical simulations were performed with parallel computing strategies using an intense computational resource allocated on the TRUBA. The simulation results in the present study revealed that the efficiency of the conventional design can be improved by 12.47% when the optimised design is implemented in the present tank.
End-user: DSTech
HPC infrastructure provider: TUBITAK ULAKBIM TRUBA HPC Center
Success story details: http://eurocc.truba.gov.tr/?page_id= 6259&lang=en
Keywords: Industrial water treatment, computational fluid dynamics, OpenFoam, HPC
ERIKTRONIK: The multiphysics experiments of the Weather Research and Forecasting Model (WRF) on precipitation patterns of Turkey
The ErikTronik company and the NCC collaborated on this project, where an operational forecast of hydrometeorological variables is aimed to be acquired at high accuracy.
The weather events are chaotic by nature, and any physic option may be time and domain-specific thus having some uncertainties. Therefore, the physics combinations doing a good job are hired as an ensemble to deal with these uncertainties. A sensitivity test can reveal the ensemble members of the multiphysics combinations. The project’s sensitivity tests were completed over the Turkey domain for 2020, with a 60-combination of model physics in 4-km resolution. The combination number, resolution, and simulation time are rather comprehensive for such sensitivity tests. Valuable information was obtained about which multiphysics ensemble responds favourably to the Turkey precipitation characteristics. The outcomes pave the way for improving weather forecasts and climate predictions over Turkey.
End-user: Eriktronik
Domain expert: Middle East Technical University
HPC infrastructure provider: TUBITAK ULAKBIM TRUBA HPC Center
Success story details: http://eurocc.truba.gov.tr/?page_id= 6271&lang=en
Keywords: Weather forecasting, earth science, HPC
PARABOL: Public Transport Analysis on HPC Infrastructure
Parabol has been carrying out R&D activities in the intelligent transportation systems sector since 2011. Currently, it performs spatiotemporal processing and analysis on the Public Transport (PT) data, around 10M rows per month. Moreover, the PT accessibility analysis should be performed by finding activity regions for each commuter. Creating value from such big data to understand the mobility patterns involves 3D (space, time, and commuter) analysis which is time-consuming and is challenging. Hence, the spatiotemporal clustering algorithm for PT user activities (STCAPT) needs parallel execution. Considering the fact that STCAPT requires a large amount of memory to run, which an HPC environment can provide, the company worked on developing an approach to run STCAPT on HPC. They developed a pipeline which was ported to Apache Spark and executed successfully. While the analysis cannot be completed reasonably on a typical server within the company data center, it took just minutes to run it with a test dataset on TRUBA. On a Spark cluster created with only three nodes of TRUBA, the test runs resulted in a 90% decrease in runtime.
End-user: Parabol
Domain expert: Middle East Technical University
HPC infrastructure provider: TUBITAK ULAKBIM TRUBA HPC Center
Success story details: http://eurocc.truba.gov.tr/?page_id=6338&lang=en
Keywords: Transportation, smart city, high-performance data analytics, big data, Apache Spark
