Doctor's Theses (authored and supervised):
"Methods for Hybrid Modeling and Simulation-Based Optimization in Energy-Aware Production Planning";
Supervisor, Reviewer: F. Breitenecker, W. Kastner, T. Pawletta;
Institut für Analysis und Scientific Computing,
oral examination: 2020-06-16.
Today, many software tools in industrial engineering employ mathematical modelling and dynamic simulation to support complex decision-making processes. In combination with heuristic or metaheuristic optimization methods, simulations enable to evaluate different planning scenarios in a systematic and iterative manner by quantifying their fitness for a given optimization target. This simulation-based approach allows to consider more complex systems than conventional analytical models, thereby offering more accurate predictions and overall improving planning quality. In this context, simulation models can be used to quantify the impact of different production schedules on the overall energy demand of a production facility, thus providing decision support to improve energy efficiency. Energy efficiency has become an increasingly important topic in industrial engineering in recent years due to their significant potential for economic savings as well as reducing ecological impact.
Yet, in order to be able to perform comprehensive energetic investigations within industrial facilities, the simulation models have to capture the complex interactions with sufficient level of detail. The models have to incorporate aspects from different engineering domains, in particular production machinery with its material flow, energy infrastructure, logistics and building physics. While material flow entities can intuitively be modelled using discrete-event methods, energy flow (including transient effects) is best described using time-continuous dynamics with differential equations. Integrating discrete and continuous modelling methods as part of a hybrid modelling and simulation approach remains a challenging task. Only few publications so far focus on hybrid systems in the context of production simulation.
Hybrid simulation in practice requires coupling discrete-event methods with differential equation solvers in a way that is not only efficient (in terms of runtime) but also formally sound (in order to produce accurate results). One common approach is to couple different simulation environments as part of a so-called co-simulation. As a drawback, the user is forced to split the overall model into different sub-models along the boundary of discrete/continuous modelling, thereby loosing component modularity. It quickly becomes cumbersome to maintain and reuse these kinds of models.
A more promising approach in this context uses a formal model description based on an extended DEVS formalism (Discrete-Event System Specification) for hybrid systems. DEVS has a sound basis as a formal model description in the academic field; in practical industrial applications, however, they have so far had little adoption. A comparison with other co-simulations also shows the potential benefits of the DEVS approach, especially for modular component-based and hierarchical hybrid modelling as well as tighter integration of continuous and discrete model parts while maintaining potentially better performance.
As one of the major drawbacks of using DEVS-based specifications in practical ap- plication, experience shows that these formalisms are often difficult to understand for non-experts, which significantly hinders potential adoption for industrial applications. Some modelling aspects may appear counter-intuitive at first, and the generic nature of DEVS requires the user to take care of low-level implementation details that are relevant for practical application. Furthermore, coordination between ODE solver and event scheduler in a distributed setting is still an open issue in hybrid DEVS formalisms.
To overcome the obstacles mentioned above, this thesis develops a concept for building a modelling framework for hybrid discrete/continuous simulation in industrial applications that allows high-level component-based model composition and component reuse. The goal is to transform the current complex and cumbersome user workflow for hybrid modelling into an improved development process that allows more seamless integration of discrete and continuous models in a modular and efficient manner.
This can be achieved by freeing the user from low-level implementation details and masking unnecessary complexity caused by the coupling mechanisms between continuous and discrete simulation. However, reducing complexity and hiding features also implies reducing flexibility for the user, which in turn restricts possible applications.
To this end, the following research questions that are to be examined in the scope of this thesis:
1. How can hybrid discrete/continuous simulation models of industrial production systems be described in an intuitive and formally sound way that enables modular composition and reuse and reduces unnecessary complexity while still retaining sufficient flexibility in practical application?
2. How can a hybrid DEVS-based modelling formalism be improved to better support such a new development process (in particular regarding handling of low-level modelling details, etc.)?
3. How can simulation-based metaheuristic optimization solutions in industry ben- efit from hybrid simulations that offer more accurate modelling by integrating production and energy simulation?
Modellbildung / Simulation / Hybride Modelle / Produktionsplanung
Created from the Publication Database of the Vienna University of Technology.