This study investigates the feasibility of automated setup of profile wrapping processes on new products using machine learning on past setup examples. The task is characterized by high complexity of the considered production system in combination with highly varying products and a very small available database. This database also reveals ambiguous ground truth due to human, unsystematic preferences. A simple geometric-physical motivated preprocessing is proposed. On the resulting data, a Deep Convolutional Neural Network in the form of an autoencoder is shown to be very suitable for predicting wrapping actions for new products. The good but improvable results are discussed extensively with respect to the technological background and possible solutions are proposed.

In this Paper a methodical approach to hybrid modelling for contextual anomaly detection on time-series data is presented. It enhances widely used proximity-or distribution-based anomaly detection approaches for industrial processes by a hybrid model. This hybrid model consists of a physical model using a priori process knowledge and a data driven model constructed through a machine learning method. The main advantage of the novel approach is that it is capable of detecting contextual anomalies that remain otherwise undiscovered.

367

Hot forming is a widely used manufacturing process of crash-relevant structural components with complex geometries. In this paper a previously presented method of anomaly detection is further optimized allowing more data sources to be used in the outlier evaluation process. The method is based on a hybrid model consisting of a physical first-principles model of the hot forming press and a neural network in series. It allows a wide range of sensor data to be considered while keeping the anomaly detection process physically explainable.

383

311

Hot forming is a widely used manufacturing process of crash-relevant structural components with complex geometries. In this paper a method is presented to detect anomalies during the hot forming process giving indications of possible quality defects. The method is based on a physical model of the thermal energy transfer between hot blank, pressing tool and cooling water. The model is built using lumped heat capacities and virtual heat resistances between the components. Within this model the temperature profiles of these components are simulated using input data from real process sensors. The physical model is then combined with a neural network to form a hybrid model. The neural network is trained using the input data and a parameter optimization algorithm and determines the currently optimal parameters of the physical model. During the production, a comparison of the simulated and the real temperature profile reveals anomalies in the hardening process. This way indications for potential quality defects are gained.

Interactive driving simulation has become a key technology to support the development and optimization process of modern vehicle components and driver assistance systems both in academic research and in the automotive industry. However, the validity of the results obtained within the virtual environment depends essentially on the adequate reproduction of the simulated vehicle movements and the corresponding immersion of the driver. For that reason, specific motion platform control strategies, so-called Motion Cueing Algorithms (MCA), are used to replicate the simulated accelerations and angular velocities within the physical limitations of the driving simulator best possible. In this paper, we present the design and evaluation of a subjective comparison of three different filter- and optimization-based MCA resulting from previous research. For that purpose, a Human-in-the-Loop experiment was conducted with 27 participants in four typical driving situations, using a hybrid kinematics motion system as an application example. The statistical analysis of the study proves that the optimization-based algorithm is preferred by the subjects regardless of the presentation sequence and the respective driving maneuver, while the filter-based approaches differ only insignificantly in their ranking. Results thus correlate with the findings of an objective evaluation of the control quality and identify further potentials for improving the driving experience.

386

For test rigs with multiaxial excitation of the specimen, the realization of a Hardware-in-the-Loop (HiL) simulation is a challenging task. In general, the excitation unit is a serial or parallel kinematic manipulator and the dynamic properties of the specimen vary in different spatial directions. Hence, the contact situation between the manipulator and the specimen requires extensive consideration. System instabilities and damage are possible. In this paper, it is demonstrated how indirect force controlled manipulators are used to provide realistic and safe HiL simulations. In this context, a HiL controller is developed for a highly dynamic vehicle axle test rig with a hydraulically actuated hexapod used as the excitation unit. The models of the HiL subsystems are shown first, and afterwards the design of the HiL controller is presented. A theoretical system analysis is provided, which includes the investigation of stability, bandwidth limitations and disturbance rejection. Simulation results are provided to emphasize the high quality of the HiL simulation.

In the area of product-based control of manufacturing processes, the potential to integrate intelligent products in process control has been recognized several times. However, the integration of greatly varying intelligent products on the job scheduling and networked control of distributed cell-based manufacturing processes have not been investigated. This paper begins with definitions of product-intelligence and a so called distributed data and interface model. Based on this, a cell-based manufacturing process with networked control architecture with different intelligent products as a function of varying product properties is presented. In addition, the product-intelligence and a highest response ratio next based job scheduling approach for flexible and reconfigurable manufacturing systems is presented. Through simulation results from a flexible candle manufacturing demonstrator scenario, it has been shown that the extended highest response ratio next scheduling approach represents a further module in the realization of product-intelligence based cyber-physical production systems with variable lot sizes and product-individual requirements.

Hybrid multi axes kinematics with kinematic redundancy have become more important in industrial applications since they offer a large workspace, high stiffness, and accuracy. However, the kinematic equations are complex because the number of solutions for the inverse kinematic problem is infinite. Therefore, a model predictive approach using optimization algorithms based on the (inverse) kinematic problem is beneficial since exisiting ambiguities can be exploited to achieve a certain goal, e.g. fast or smooth movement. However, in order to be used in an industrial context, it is crucial, that the approach is computationally efficient so that it can be implemented on a Programmable Logic Controller (PLC). This paper presents a computationally efficient model predictive feedforward compensation for position control of a multi axes hybrid kinematic. The simulative results show that a system with such a model predictive approach reaches set-points faster than current approaches. Furthermore, we focus on the future implementation of the algorithms on a PLC.

In the area of product-based control of manufacturing processes, the potential to integrate intelligent products in process control has been recognized several times. However, the effect of greatly varying intelligent products on the control and hybrid system dynamics of the individual (sub-) processes have not been investigated. This article begins with a definition of product-intelligence and a distributed data and interface model. Based on this, a decentralized hybrid reconfiguration structure with different intelligence-based reconfiguration strategies as a function of varying product properties is presented, using the example of a conveyor belt system. Through the implementation on a industrial controller, it has been shown that the concept represents a further module in the realization of cyber-physical systems through comprehensive analysis and synthesis of a highly flexible hybrid control loop.

The Autonomous Intersection Management (AIM) will be a future method for the Intelligent Transportation System. It combines wireless communication and the autonomous vehicle in order to create the new concept for managing road traffic more safely and efficienly. The distributed control principle is applied to the intersection network to control the traffic in the macroscopic level. The Vehicle to Infrastructure (V2I) and Infrastructure to Infrastructure (I2I) communication are used to exchange the traffic information between a single autonomous vehicle to the network of autonomous intersections The discrete time consensus algorithm is implemented to coordinate the gross traffic density of an intersection and its neighborhoods in the network. The boundary condition for the uncongested flow is created by using the Greenshield's traffic model. The proposed method represents the ability to maintain the traffic flow rate of each intersection and operates with the uncongested flow condition. The simulation results of the network of a multiple autonomous intersection are provided.

In this work we present an approach for the optimization of distributed test rigs that are coupled only by information processing. Our application examples are an active suspension system and a linear drive with an active air gap adjustment which both represent a module of the rail-bound vehicle RailCab. Hierarchical optimization is used to combine the module-related optimal operating strategies, which are based on two distinct multiobjective optimizations. A Pareto front of the entire system is presented as a result of the hierarchical optimization.

In this paper, a new model based optimal motion controller for driving simulators is presented. Models of human perception systems in combination with a simulator dynamic model are used in the design of the optimal controller. The basic idea of the new approach is based on minimizing the difference between the perceived signals in the vehicle and the perceived signals in the simulator. In addition, a new approach based on the human vestibular system is used to represent the sustained acceleration for the tilt coordination. Compared to conventional motion cueing algorithms, the proposed approach provides a more realistic impression, the workspace of the simulator is better exploited and all the constraints of the driving simulator are always respected. Furthermore, a new objective quality criterion is suggested to evaluate the new approach. A comparison between the classical washout, the adaptive washout and the new optimal control strategy is carried out.