Mirko Kück

Hybrid approach for the integrated scheduling of production and transport processes along supply chains

The rise of new information and communication technologies leads to enhanced information transparency in supply chains. In order to utilise the resulting potentials, novel scheduling approaches that are capable of processing large amounts of data and coping with dynamic disturbances of manufacturing and transport stages have to be developed. For this purpose, the paper at hand proposes a hybrid approach for the integrated scheduling of production and transport processes along supply chains. The procedure combines mixed integer linear programming, discrete event simulation and a genetic algorithm. Obtained results show a significant reduction in the number of late orders, substantiating that proper scheduling approaches combined with information visibility allow for operational improvements in manufacturing supply chains.

A Genetic Algorithm to Optimize Lazy Learning Parameters for the Prediction of Customer Demands

The prediction of time series is an important task both in academic research and in industrial applications. Firstly, an appropriate prediction method has to be chosen. Subsequently, the parameters of this prediction method have to be adjusted to the time series evolution. In particular, an accurate prediction of future customer demands is often difficult, due to several static and dynamic influences. As a promising prediction method, we propose a lazy learning algorithm based on phase space reconstruction and k-nearest neighbor search. This algorithm originates from chaos theory and nonlinear dynamics. In contrast to widely used linear prediction methods like the Box-Jenkins ARIMA method or exponential smoothing, this method is appropriate to reconstruct additional influences on the time series data and consider these influences within the prediction. However, in order to adjust the parameters of the prediction method to the observed time series evolution, a reasonable optimization algorithm is required. In this paper, we present a genetic algorithm for parameter optimization. In this way, the prediction method is automatically fitted accurately and quickly to observed time series data, in order to predict future values. The performance of the genetic algorithm is evaluated by an application to different time series of customer demands in production networks. The results show that the genetic algorithm is appropriate to find suitable parameter configurations. In addition, the prediction results indicate an improved forecasting accuracy of the proposed prediction algorithm compared to linear standard methods.

Potentials of Nonlinear Dynamics Methods to Predict Customer Demands in Production Networks

Nowadays, markets are characterized by increasing dynamics and complexity. In particular, customer demands are often highly volatile. These conditions complicate demand forecasting and reduce the average accuracy of forecasting data. Nevertheless, manufacturing companies have to predict customer demands precisely, in order to achieve a well-founded production planning and control. The paper at hand deals with methods to predict customer demands in application scenarios of production logistics. Firstly, forecasting methods for smooth customer demand are described with a particular emphasis on nonlinear dynamics methods. Subsequently, a new algorithm to predict intermittent demand is introduced. In both cases of demand evolution, different methods are applied to predict demand data generated by a discrete-event simulation of a production network. Forecasting results are interpreted and the different methods are rated regarding their applicability. The research displays that an application of nonlinear dynamics methods can lead to improved forecasting accuracy.

Potential of data-driven simulation-based optimization for adaptive scheduling and control of dynamic manufacturing systems

The increasing customization of products, which leads to greater variances and smaller lot sizes, requires highly flexible manufacturing systems. These systems are subject to dynamic influences and demand increasing effort for the generation of feasible production schedules and process control. This paper presents an approach for dealing with these challenges. First, production scheduling is executed by coupling an optimization heuristic with a simulation model. Second, real-time system state data, to be provided by forthcoming cyber-physical systems, is fed back, so that the simulation model is continuously updated and the optimization heuristic can either adjust an existing schedule or generate a new one. The potential of the approach was tested by means of a use case embracing a semiconductor manufacturing facility, in which the simulation results were employed to support the selection of better dispatching rules, improving flexible manufacturing systems performance regarding the average production cycle time.

Meta-learning with neural networks and landmarking for forecasting model selection an empirical evaluation of different feature sets applied to industry data

Although artificial neural networks are occasionally used in forecasting future sales for manufacturing in industry, the majority of algorithms applied today are univariate statistical time series methods for level, seasonal, trend or trend-seasonal patterns. With different statistical methods created for different time series patterns, large scale applications on 10,000s of times series require automatic method selection, often done manually by human experts based on various time series characteristics, or automatically using error metrics of past performance. However, the task of selecting adequate forecasting methods can also be viewed as a supervised learning problem. For instance, a neural network can be trained as a meta-learner relating characteristic time series features to the ex post accuracy of forecasting methods for each time series. Past research has proposed different sets of time series features for meta-learning including simple statistical or information-theoretic as well as model-based features, but have neglected the use of past forecast errors. This paper studies the predictive accuracy of using different feature sets for a neural network meta-learner selecting between four statistical forecasting models, introducing error-based features (landmarkers) and statistical tests as time series meta-features. A large-scale empirical study on NN3 industry data shows promising results of including error-based feature sets in meta-learning for selecting time series forecasting models.