A runtime adaptation concept to reinforce versatility in industrial automation
Elfahaam, Haitham Ahmed; Epple, Ulrich (Thesis advisor); Vogel-Heuser, Birgit (Thesis advisor)
Aachen : VDI Verlag GmbH (2019, 2020)
Book, Dissertation / PhD Thesis
In: Fortschritt-Berichte VDI. Reihe 8, Mess-, Steuerungs- und Regelungstechnik 1267
Page(s)/Article-Nr.: 1 Online-Ressource (XII, 116 Seiten) : Illustrationen, Diagramme
Dissertation, RWTH Aachen University, 2019
In the process control engineering domain, various initiatives around the world (e.g. “Industry 4.0” in Germany) play a crucial role in directing the research and development. In the new generation of industrial automation, a new architecture is introduced where the communication hierarchy of the automation pyramid is dissolved in order to increase the flexibility of the production systems. One of the objectives of this architecture is to achieve “Adaptability” or in other words to enable industrial plants to react to unplanned changes. Furthermore, design principles like decentral decision making and interconnectedness are widely promoted. In order to achieve the aforementioned goals, various new functionalities (e.g., Self-X functionalities and optimizations) and information (e.g., asset administration shell) are being introduced to the current production systems which did not exist in the conventional ones thus causing a dynamic overhead to the available resources (computation, communication, dynamic memory, etc.). In the conventional systems, during the engineering phase, control logic and functionalities are designed and then deployed to the computation nodes in the automation network. In some cases, an optimized distribution profile for the loads are computed prior to the initial deployment and accordingly the load is distributed amongst the network endpoints. However, the dynamic aspect of the load variations introduced in the newly introduced automation paradigm is not taken into consideration. System adaptation to the varying loads is required to readjust the loads and balance the resources consumption in the network. In industrial automation, safety aspects play a crucial role. Hence, a prerequisite for this framework is to not compromise the stability of the production system. The objective of this dissertation is to establish a framework for a seamless integration of a deployment platform that can, through redeployment and adjustment of software components, balance the resources consumption overhead amongst the automation network participants, establish redundancy of the different components, improve the communication quality of service and adapt the system according to the rapid and dynamic changes imposed. Thorough analysis and investigations for stability and the production system dynamics are conducted. The goal of these investigations is to ensure that the introduced framework does not affect the performance in any undesired manner, e.g., causing the loads to oscillate in the network or affecting the system performance with a non converging redeployment processes of the software components. Hence, additional to these investigations, a multi optimization criteria load balancing model is constructed to investigate the behavior or multidimensional optimizations. Moreover, performance enhancements analysis is conducted through investigating the automation networks and constructing regression models to compute the optimal parameters for load redeployment. Furthermore, a prototype implementation to reinforce the presented concepts and validate the conducted investigations is realized. The prototype considers an aluminum cold rolling mill use-case and utilizes two different approaches namely decentral algorithms and agent systems approaches to perform the load balancing from two different perspective namely resources and component perspectives respectively. In the former approach, the algorithm uses mathematical formulas (e.g., total square error) to compute the optimum load balancing profile from a decentral perspective and cooperates with other network participants to achieve the optimum load distribution profile on a global scale. On the other hand, in the latter approach, the components are considered as independent agents that wander the network. The information incubated within an agent is used (e.g. optimal routed path according to a given recipe) to anticipate the load distribution in the network and thus adjust the placement of the components (agents) accordingly. The presented prototype implementation uses the runtime environment ACPLT/RTE and acts as extension library to provide the load balancing functionalities. The implementation uses the demonstrator from the SMS-group that simulates a cold rolling mill plant.