Benjamin Klöpper

Probabilistic planning integrated in a multi-level dependability concept for mechatronic systems

Self-optimizing mechatronic systems are a new class of technical systems. On the one hand, new challenges regarding dependability arise from their additional complexity and adaptivity. On the other hand, their abilities enable new concepts and methods to improve the dependability of mechatronic systems. This paper introduces a multi-level dependability concept for self-optimizing mechatronic systems and shows how planning can be used to improve the availability and reliability of systems in the operating stages.

Simulation Assisted Optimization and Real-Time control Aspects of Flexible Production Systems Subject to Disturbances

Several types of production systems have been studied and researched in the past using either simulation andor optimization methods. In this paper we describe the design and development of a simulation assisted predictive-reactive system for scheduling and rescheduling a typical flexible production system configuration. Aspects like the combined use of simulation and optimization to solve complex scheduling and rescheduling tasks are described in view of system stability and some of the broader production system elements like buffer sizing and material handling equipment. Results show that combining simulation and optimization for predictive scheduling resulted in better and valid performance measures for a typical example. Results also show that some newly addressed aspects of stability and real-time control can be handled efficiently using a combination of simulation and optimization. Our discussions only bolster the claim that simulation is an indispensable tool in managing complex production systems

Aspects of Agent Based Planning in the Demand Driven Railcab Scenario

Multi-agent Systems are a promising approach to implement flexible and effective decentralized control mechanismen for the logistic domain. This paper intends to introduce the key features of a multi-agent system that is under development for several years. The agents autonomously plan, optimize, and control a railway transportation system. For this purpose the agents interact, communicate and negotiate. This paper presents an overview how different techniques from operations research, artificial intelligence and soft computing are integrated into a multi-agent system to solve this difficult and complex real life problem.

A Reinforcement Learning Approach for Learning Coordination Rules in Production Networks

In production networks companies need fast reactions due to changes of supply and demand. To realize such a change management in an effective way the involved companies have to synchronize their quantities and capacities collaboratively. For these purposes the multiagent system MASCOPP was developed at the Heinz Nixdorf Institute, which tries to eliminate conflicts in a production network, based on changes of plans, through bilateral communication between the involved companies. Human experts have to configure the system by creating coordination rules to solve the conflicts. In this paper we introduce a machine learning concept to learn these coordination rules objectively by a reinforcement learning approach.

Bi-objective-optimisation of an international transportation problem for CO 2 -efficient schedules

The transportation sector accounts for a considerable portion of global CO2 emissions and the demand for transport, especially for international transport, is increasing. Current approaches to reduce CO2 emission on an operational level ignore time-related objectives and focus on simultaneous optimisation of economic and ecological objectives. In general, considering the minimisation of CO2 emissions in transportation scheduling introduces a trade-off between economic and ecological concerns. The concept of eco-efficiency is important in order to evaluate the economical value and ecological quality of business decisions. Nevertheless, this concept has primarily been applied to strategic decisions. In the present paper we demonstrate the applicability of this concept to operative decisions and introduce a bi-objective evolutionary optimisation approach for the approximation of CO2-efficient schedules in an international transportation problem.

A modelling approach for dynamic and complex capacities in production control systems

In this paper we introduce some aspects of the development process of a production planning tool for a leading European car manufacturer. In this project we had to face a gap between theoretical problem definition in manufacturing planning and control and the actual requirements of the dispatcher. Especially the determination of production capacities and product processing times was a severe problem. The software system had to support the derivation of these important variables from shift plans, factory calendars and exceptional events. In order to implement this we defined a formal model that is a combination of the theoretical and practical view on manufacturing problem. The Model for Serial Manufacturing allows the dispatcher to provide up-to-date information in an easier way then provided by standard ERP systems about the production systems and transfers the information automatically to planning algorithms. Thus the production planning and control is always performed on the most recent information.

The Shortest Path Problem in Uncertain Domains -- an Agent based Approach with Bayesian Networks

Real world problems, e.g. from transport domain, are typically non-deterministic and uncertain. Although many approaches, especially form the area of operations research assume constant parameters like ride duration or availability of connections in networks, these parameters depend on a large number of influences and are not constant. The paradigm of Bayesian thinking teaches us, that we can use partial knowledge of the influences to derive better estimations of these variables. In this paper, we introduce a multiagent system (MAS) that implements a distributed hybrid Bayesian network and is able to create a dynamic probabilistic model of the shortest path problem, in order to create better estimation and thereby better plans

Risk Averse Shortest Path Planning in Uncertain Domains

Real world problems, e.g. from transport domain, are typically non-deterministic and uncertain. Although there are some approaches, which try to forecast uncertain parameters like travel time, the uncertainty is rarely included in the planning process. In this paper a probabilistic forecasting method for travel time in a railway network is introduced which considers the dependencies between decisions during the planning process. The information provided by forecasting is used to develop a risk averse shortest path algorithm which minimizes the risk of delay.

A knowledge-based giffler-thompson heuristic for rescheduling job-shops

Even today rescheduling in job-shop systems is still a challenge. There are approaches to solve the problem like analytical, heuristic and simulation ones. Analytical methods cannot meet the requirements of rescheduling regarding solution time, especially for large problem instances. Analytic approaches as well as simulation based systems need a long calculation time. To generate good solutions a lot of simulation runs have to be made. Thus, extensive research was done in heuristic rescheduling systems. Usually, dispatching rules are used. Their drawback is that it is impossible to define a superior dispatching rule for all situations in the workshop. To solve this problem, we intend to combine the Giffler/Thompson heuristic with a knowledge based system - a naïve bayes classifier with offline generated training data. This combination enables the selection of the best dispatching rule dynamically, depending on the systems state. The paper presents the concept of our approch; a first stage of research progress. Therefore, preliminary results on learning scheduling decisions are shown.

An Agent Based Modelling Approach for Stochastic Planning Parameters

Many planning problems are influenced by stochastical environmental factors. There are several planning algorithms from various application domains which are able to handle stochastic parameters. Correct information about these stochastic parameters has impact on the quality of plans. There is a lack of sufficient research on how to obtain this information. In this paper, we introduce a Multiagent System (MAS) that is able to model stochastic parameters and to provide up-to-date information about these parameters. Due to their access to locally available informations expert agents are used, which apply the paradigm of Bayesian Thinking in order to provide high quality information to planning agents.