Simulation Science

Autonomous vehicles (=AV) enabling driverless transport may change the ways of traveling and traffic volumes dramatically. To estimate potential impacts of AV on traffic in an urban area nine scenarios are examined, varying the rate of carsharing, ridesharing and the availability of rail services. The number of required vehicles, vehicle kilometers and the necessary number of parking spaces quantify each scenario.

The study builds on an existing travel demand model of the Stuttgart Region. An algorithm extends this model for bundling person trips in ridesharing systems and by an algorithm for vehicle blocking. The results show that the size of the car fleet can be reduced considerably. The vehicle kilometers traveled in the network, can only be reduced in cases where most travelers use ridesharing instead of carsharing or privately owned cars. However, an increase of the car kilometers traveled is more likely and may lead to a lower quality of traffic flow.

We investigate the so called Double Row Facility Layout Problem (DRFLP). Given a set of departments with given lengths and pairwise transport weights between them, the aim is to assign the departments to two rows such that the weighted sum of the distances between them is minimized and such that the departments do not overlap. The DRFLP is known to be rather challenging. Even with the best approach known in literature, which is based on an enumeration over all row assignments of the departments and where only the center-to-center distances are measured, the largest instance solved to optimality contains only 16 departments. In this paper we show how the existing models can be extended in various directions in order to handle more aspects that are important in real-world applications such as vertical distances between the departments and restricting the size of the layout area. We also show how the structure of real-world instances, which often contain several departments of the same type, can be exploited in mathematical optimization. This allows us to solve a realistic instance with 21 departments in reasonable time. Furthermore, we propose a new approach which combines optimization and simulation. Here simulation allows the evaluation of the optimized solutions with respect to several performance indicators which play an important role for a smooth production apart from the weighted transport distances. If problems are detected, this information is included in the mathematical models by extending these.

In this paper, we introduce the MuRO-NIMBUS method for solving multiobjective optimization problems with uncertain parameters. The concept of set-based minmax robust Pareto optimality is utilized to tackle the uncertainty in the problems. We separate the solution process into two stages: the pre-decision making stage and the decision making stage. We consider the decision maker’s preferences in the nominal case, i.e., with the most typical or undisturbed values of the uncertain parameters. At the same time, the decision maker is informed about the objective function values in the worst case to support her/him to make an informed decision. To help the decision maker to understand the behaviors of the solutions, we visually present the objective function values. As a result, the decision maker can find a preferred balance between robustness and objective function values under the nominal case.

In the last two decades, water consumption in Germany has been decreasing which causes water tanks and pipes in water distribution systems to work inefficiently. This paper presents a mathematical optimization model to optimize water tanks in a water distribution system. Due to the hydraulic properties in water distribution systems the model is a non-convex Mixed Integer Quadratically Constrained Program (MIQCP). For problem instances of realistic size, the model cannot be solved within reasonable time with exact solution methods. We use different heuristic solution methods to solve the problem, such as a Simulated Annealing (SA) algorithm, a Shuffled Complex Evolution (SCE) algorithm as well as a Shuffled Frog-Leaping Algorithm (SFLA). These methods are combined with a hydraulic simulation to evaluate the solutions. The results of each method are compared to an exact solution method and discussed in this paper.

The simulation of dry particle packings and their geometrical properties is of great importance to material sciences. Substantial acceleration of the simulation can be obtained using parallel hardware (GPU), but this requires specialized data structures and algorithms. We present a parallel version of the so-called collective rearrangement algorithm that allows to simulate random close packings of up to several million spherical particles from an arbitrary particle size distribution. The empirical time complexity of our implementation is almost linear in the number of spheres.

A parallel hybrid Monte Carlo/Molecular Dynamics coupled framework has been developed to overcome the time scale limitation in simulations of segregation of interstitial atoms in solids. Simulations were performed using the proposed coupling approach to demonstrate its potential to model carbon segregation in ferritic steels with a single dislocation. Many simulations were carried out for different background carbon concentrations. This paper is a first step towards understanding the effect of segregation of interstitial atoms in solids and its influence on dislocation mobility in external fields. To this end, we carried out MD simulations, where shear forces were applied to mechanically drive screw dislocation on configurations with segregated carbon atoms. The results are compared with a reference system containing homogeneously distributed carbon atoms where the influence of segregated carbon on dislocation mobility could be observed. Simulation results gave qualitative evidence that the local concentration of interstitial solutes like carbon provides a significant pinning effect for the dislocation.

The microstructure of lithium-ion battery electrodes has a major influence on the performance and durability of lithium-ion batteries. In this paper, an overview of a general framework for the simulation of battery electrode microstructures is presented. A multistep approach is used for the generation of such particle-based materials. First, a ‘host lattice’ for the coarse structure of the material and the placement of particles is generated. Then, several application-specific rules, which, e.g., influence connectivity are implemented. Finally, the particles are simulated using Gaussian random fields on the sphere. To show the broad applicability of this approach, three different applications of the general framework are discussed, which allow to model the microstructure of anodes of energy and power cells as well as of cathodes of energy cells. Finally, the validation of such models as well as applications together with electrochemical transport simulation are presented.

Based on virtual materials testing, which combines image analysis, stochastic microstructure modeling and numerical simulations, quantitative relationships between microstructure characteristics and effective conductivity can be derived. The idea of virtual materials testing is to generate a large variety of stochastically simulated microstructures in short time. These virtual, but realistic microstructures are used as input for numerical transport simulations. Finally, a large data basis is available to study microstructure-property relationships quantitatively by classical regression analysis and tools from statistical learning. The microstructure-property relationships obtained for effective conductivity can also be applied to Fickian diffusion. For validation, we discuss an example of Fickian diffusion in porous silica monoliths on the basis of 3D image data.

We aim to simulate software processes in order to predict the structural evolution of software graphs and assure higher software quality. To make our simulation and therefore the results more accurate, we need to model real world practices. In this paper, we consider the specific problem of including software refactorings in our simulation. We describe these refactorings as graph transformations and apply parameters we collected from open source projects.

Many simulations require large amounts of computing power to be executed. Traditionally, the computing power is provided by large high performance computing clusters that are solely built for this purpose. However, modern data centers do not only provide access to these high performance computing systems, but also offer other types of computing resources e.g., cloud systems, grid systems, or access to specialized computing resources, such as clusters equipped with accelerator hardware. Hence, the researcher is confronted with the choice of picking a suitable computing resource type for his simulation and acquiring the knowledge on how to access and manage his simulation on the resource type of choice. This is a time consuming and cumbersome process and could greatly benefit from supportive tooling. In this paper, we introduce a framework that allows to describe the simulation application in a resource-independent manner. It furthermore helps to select a suitable resource type according to the requirements of the simulation application and to automatically provision the required computing resources. We demonstrate the feasibility of the approach by providing a case study from the area of fluid mechanics.

Today’s companies in high-tech industries develop products of high complexity which consist of complicated subsystems with many heterogeneous components integrated together. As the system complexity increases, it becomes increasingly more challenging to manage the tedious development process. The Capability Maturity Model Integration (CMMI) was proposed as a general framework for process management and improvement which judges the maturity of a process. Simulations have long been regarded as complex and integrated systems. Simulation system engineering manages the total simulation system’s life-cycle process. The adaptation of the CMMI for simulation life-cycle processes is envisioned as a domain specific solution for simulation process management and improvement. This article investigates the opportunities of extending the CMMI engineering process area with emphasis in simulation system engineering, having its roots from IEEE Recommended Practice for Distributed Simulation Engineering and Execution Process (DSEEP).

Holistic simulation aids the engineering of cyber physical systems. However, its complexity makes it expensive regarding computation time and modeling effort. We introduce multi-level-simulation (Our Multi-Level-Simulation approach was already published in [1]. The description of our approach in this paper is based on this publication and updates it. This description is the context to the results on learning State mappings within Multi-Level-Simulations presented in this paper.) as a methodology to handle this complexity. In this methodology, the required holistic perspective is reached on a coarse level, which is linked with multiple detailed models of small sections of the system. In order to co-simulate the levels, mappings between their states are required. This paper gives an insight into the current state of progress of using well known machine learning techniques for regression to generate these mappings using small sets of labeled training data.

Evaluations of Internet of Things (IoT) and Wireless Sensor Network (WSN) applications demonstrate the significant and still existing gap between examinations with generic simulation environments and real-life (e.g., field test) or controlled (e.g., testbed) sensor network deployments in terms of realistic and accurate results. The separated use of single examination approaches is often not enough to overcome all evaluation challenges. We therefore propose a combination of discrete-event simulation, radio-channel emulation, and real hardware working together on different layers of the protocol stack of the system-under-test. Our combined approach reduces the gap between abstract simulations and network testbed experiments by providing adjustable radio conditions for repeatable evaluations of WSN and IoT networks.

In the field of software evolution, simulating the software development process is an important tool to understand the reasons why some projects fail, yet others prosper. For each simulation however, there is a need to have an assessment of the simulation results. We use Conditional Random Fields, specifically a variant form based on the Ising model from theoretical physics, to assess software graph quality. Our CRF-based assessment model works on so called Software Graphs, where each node of that graph represents a software entity of the software project. The edges are determined by immediate dependencies between the pieces of software underlying the involved nodes.

Because there is a lack of reference training data for our kind of evaluation, we engineered a special training paradigm that we call the Parsimonious Homogeneity Training. This training is not dependent on reference data. Instead of that it is designed to produce the following two effects. First, homogenizing the assessment of highly interconnected regions of the software graph, Second, leaving the assessment of these regions in relative independence from one another.

The results presented demonstrate, that our assessment approach works.

Data is often mined using clustering algorithms such as Density-Based Spatial Clustering of Applications with Noise (DBSCAN). However, clustering is computationally expensive and thus for big data, parallel processing is required. The two prevalent paradigms for parallel processing are High-Performance Computing (HPC) based on Message Passing Interface (MPI) or Open Multi-Processing (OpenMP) and the newer big data frameworks such as Apache Spark or Hadoop. This paper surveys for these two different paradigms publicly available implementations that aim at parallelizing DBSCAN and compares their performance. As a result, it is found that the big data implementations are not yet mature and in particular for skewed data, the implementation’s decomposition of the input data into parallel tasks has a huge influence on the performance in terms of run-time due to load imbalance.