At the eScience Center we can provide the support you need to solve large, advanced, complex computational problems which are too big for a single machine to compute.

Check out the below videos for more information.

The Faculty of Engineering and Science – Department of Energy Technology – AAU

The transport, agglomeration and subsequently deposition of small adhesive particles play important roles in many industrial and fundamental processes. These processes range from particles accumulating at heat transfer surfaces, particles blocking pores in membrane filtration systems, particles being inhaled and deposited in our lungs to interstellar medium agglomerating causing early stages of new planets to form in space.
The purpose of this study is look into how particle response time and particle adhesiveness affect the agglomeration rate. To obtain a fully-developed flow, the computational domain is made periodic in the stream-wise direction. To avoid imposing any limitations on the agglomeration process, the soft-sphere DEM approach is used to resolve how particles collide, agglomerate, deposit and are re-entrained into the fluid due to collisions with other particles or by fluid forces.

Publication

Faculty of Engineering – Center for Nano Optics – The Mads Clausen Institute – SDU

Plasmonics studies collective oscillations of free electrons, typically excited by light, in conducting media such as metals. While these interactions are commonly treated classically, we focus on situations where classical electrodynamics is interfacing regimes with quantum physics.

Faculty of Engineering – Maersk Mc-Kinney Moller Institute – SDU Embodied Systems for Robotics and Learning – SDU

The aim of this project is to rethink the current standard colorectal cancer (CRC) screening program and significantly improve the efficiency in terms of accuracy, acceptability, reduced complication rate and cost effectiveness.

An endoscopic camera pill produces up to 500,000 images per patient. All the recorded images are investigated manually by trained nurses which results in inefficient use of personnel. Each year, approximately 130,000 patients will go through screening. We are developing a new deep learning algorithm to fully automatise this process. ABACUS 2.0 offers the platform to handle large scale data and will be used for machine learning algorithms and big data analytics for detection of any cancer precursor, such as polyp detection, classification and characterization.

Department of Civil Engineering, Aalborg University (AAU)

I work with computational hydrodynamics for marine applications and use ABACUS 2.0 to do CFD simulations of wave energy converters that can’t be properly simulated with standard linear diffraction/radiation models. These computations are very computationally expensive so access to HPC is a must. The idea is to use the results from the CFD simulations to build nonlinear static blocks to be added to the faster linear solver, in order to approximately incorporate the highly nonlinear wave overtopping in the linear models.

Photo: Simulation of the WaveDragon wave energy device using OpenFOAMs two-phase RANS solver coupled to an in-house mooring dynamics model.

Department of Electronic Systems – Aalborg University (AAU)

My research field is cyber security, where we are mainly using machine learning algorithms to classify network traffic. Our aim is to be able to distinguish malicious traffic from benign traffic, thus allowing for detecting computers/devices infected with viruses, trojans, botnets, and other kind of malware.

The experiments where HPC helped us dealt with correlation of alarms from Intrusion Detection Systems using Neural Networks. Without HPC the training time was very high, which made it impossible for us to carry out the experiments – HPC reduced the training time by a factor of 35.

The aim of this particular work was to correlate alarms from IDS. The problem is that operators receive so many (false) alarms that it is hard to react timely and properly to true alarms. So these alarms and underlying features are what we have been running on ABACUS 2.0. The preliminary results indicate that the proposed method produces approximately 10 fold reduction in number of alerts, which significantly reduces the need for involvement of a human analysts.

Robotics, Evolution, and Art Lab- Department of Computer Science – IT University of Copenhagen

We work on designing robotic morphologies and controllers for locomotion tasks by using evolutionary algorithms. Specifically, we employ modular robots, simple and autonomous devices that can be assembled together to achieve a functional robot. In ABACUS2.0, we run physical simulations to evaluate the suitability of a robotic morphology and a controller. As these simulations are very time consuming, they are carried out in parallel in a cluster. Our results show that certain module´s geometries and generative encodings can greatly speed up the optimization process. Additionally, the best robots can be assembled and tested in reality, which give us a measurement of their performance. Comparing real and simulated results, we can shed light on how to reduce the reality gap.

Faculty of Engineering – The Mads Clausen Institute – SDU NanoSYD – SDU

Ultrasonic flow meters are used in a wide range of applications. This project specifically focuses on the particular problem of measuring flow in multi-phase flow condition, where the flow media consists of two or more substances. One industrially relevant example is the so-called bubbly flow scenario, where the flowing media is a gas/liquid mixture. In state-of-the art flow meters, this inhomogenity significantly affects the measuring accuracy, since ultrasonic signals get scattered from the second phase, adding distortions to the received signal. In practice, this leads to highly error-prone, unreliable measurement results. In this project in collaboration with Siemens Flow Instruments, we develop a numerical model that helps better understand and predict measurements in this screnario, with the ultimate goal of overcoming the described problem.

As it is nearly impossible to establish a measurement setup, where multiphase flow conditions would be sufficiently controlled to compare a single measurement to single simulation, we have to introduce a statistical modelling approach. Series of simulations hence need to performed for various phase distributions, flow speeds and flow profiles, in order to look for common received signal signatures. To this end, we implement our model in Matlab and run this vast number of necessary simulations in a highly parallel fashion on the ABACUS 2.0 supercomputer.

Faculty of Engineering – Maersk Mc-Kinney Moller Institute – SDU Embodied Systems for Robotics and Learning- SDU

The research efforts we address with ABACUS 2.0 are related to data volume. The variety of sensor installed in wind turbines offer the unique opportunity to use large scale data to predict events altering the wind turbines’ performance. The problem at hand is the limited abilities of handling those data by local workstations. ABACUS 2.0 offers the platform to handle large scale data and perform analysis within reasonable time compared to human efforts. With ABACUS 2.0 we are turning various data types (times-series, logs, and other sensory data) into valuable insight, such as the remaining lifetime of turbines. Recent results have shown that learning from large scale wind turbine data can provide prediction horizons of several month prior to a failure of a wind turbine. In the future we want to utilize the parallel computing power to decrease the dimension of wind turbine sensor data into more informative data series. Ultimately obtaining knowledge about the underlying root cause for failures in wind turbines.

Faculty of Engineering – The Mads Clausen Institute – SDU NanoOptics – SDU

At SDU Nano Optics, we utilize the strong interaction of light with nanostructured components to conduct research within the fields of quantum plasmonics and optical metasurfaces, where the accurate description of the systems requires a simultaneous knowledge of the distribution of the electromagnetic field at both the nano- and micrometer scale, which makes it a computationally demanding task. The calculations performed on ABACUS 2.0 are all depending on the finite element approach (FEA) implemented in the commercial software Comsol Multiphysics. Based on the simulation results, we have demonstrated various optical functionalities, such as focusing, beam deflection, integrated single-photon source and surface plasmon polariton couplers.