Study On Topology Design And Optimization Of Industrial Ethernet Networks

With the increasing complexity of industrial networks at field level, network topology design with quality of service (QoS) guarantee becomes a challenging task for engineers. As network complexity increases, established network design methods based on engineering rules-of-thumb cannot achieve high efficiency and reliability. Previous research of general network topology design did not consider the particular requirements of industrial applications. Hence, it becomes an import topic to systematically study the topology design and optimization of industrial networks to provide network QoS guarantee.Focussing on industrial Ethernet, i.e. the technology forming the basis of the new generation of industrial networks, this thesis investigates methods to guarantee the network performance of industrial communication networks by topology design, based on the consideration of the specific requirements of industrial applications, the QoS related features of industrial Ethernet protocols, and the increasing network complexity in the industrial field. The achievements of this thesis can be summarized as follows.1. A set of fast methods for real-time and reliability evaluation in industrial Ethernet networks are developed. To this aim, an improved upper bound of the end-to-end transmission delay is formulated using network calculus theory by considering the hardware limitations of network nodes. The reliability of basic topologies defined in IEC 61784-1 is formulated. A real-time risk indicator is proposed to evaluate the degration of real-time performance in case of failures. The simulation results show that the proposed indicator effectively evaluates the robustness of real-time performance of multi-ring topologies without the need to evaluate the real-time perforamence in all possible failure sitiuations.2. For industrial applications with network structure constraints, the device allocation problem in industrial Ethernet networks is analyzed based on a novel metric, the relative delay. Evaluation of several test cases shows that the relative delay can adequately indicate whether the network real-time performance satisfies the diverse real-time requirements of traffic flows in industrial applications, and leads to a more effective search during the execution of network optimization algorithms. Furthermore, three distinct algorithms for optimized device allocation are developed and evaluated:ssGA (steady state genetic algorithm), rVNS (reduced variable neighborhood search), and a hybrid ssGA-rVNS algorithm based on an adaptive neighborhood transfer strategy. All three algorithms improve the efficiency compared to the algorithm proposed in previous works under different network sizes and traffic distributions. The hybrid ssGA-rVNS algorithm exhibits the best performance among the three proposed schemes. Even when the nework size gets large, the real-time QoS can be guaranteed in the solutions obtained by the hybrid ssGA-rVNS algorithm while the computation complexity remains moderate.3. For the industrial application without network structure constraints, a more general topology design approach for industrial Ethernet networks is described with the following features. First, a co-design of physical topology and logical topology is considered to optimize both real-time performance and reliability. Second, topologies are constrained to combinations of rings to support ring redundancy management protocols. Two multi-objective evolutionary algorithms, cNSGA-II (constrained Nondominated Sorting Genetic Algorithm-II), and MOVNS (Multi-objective Variable Neighborhood Search) are developed for this problem. Problem specific representations of potential solutions are proposed for both algorithms to transfer this Bilevel programming problem to a single level optimization problem. The evaluation of several test cases shows that 1. the proposed real-time risk indicator is effective as a real-time optimization criterion for the topology design,2. both algorithms can improve the performance compared to the existing genetic algorithm at a lower computational effort,3. cNSGA-II algorithm greatly decreases the computation time compared to the genetic algorithm proposed in previous works, and provides reasonably good solutions for the topology design problem in industrial Ethernet networks,4. MOVNS algorithm further improves the algorithm efficiency compared to cNSGA-II algorithm, and is able to obtain a Pareto solution set with better quality and diversity. The design methods effectively reduce the network cost and improve QoS performance in all considered test scenarios.Through the study of the performance evaluation methods for industrial Ethernet networks, the formal description of the topolology design problem for industrial Ethernet networks and the customized optimization algorithms proposed in this thesis, a comprehensive topology design methodology has been presented in this paper that assists control engineers during the system design phase in designing topologies of industrial Ethernet networks with network QoS guarantee.