Multilayer Network-Based Production Flow Analysis
Read this article. The intent is to explore production technologies in relation to flow analysis. Pay attention to how production flow analysis is defined. Do you agree or disagree?
Conclusions
A multilayer network model was developed for production flow analysis to represent the physical and functional domains of production systems by taking into account the aspects of the structure of the system, the variety of machines, products, components, and operators and their interdependencies.
Most of the layers of the model are represented by a bipartite graph, where edges represent material, energy, or information flows and attributes of the objects represented by the nodes of the graph. It was highlighted that the nodes and connections could be easily defined based on standards of process management. As the layers of the network represent different aspects of the production system, the proposed model is flexible and easily extendable.
Following the introduction of the new modeling concept, it was demonstrated how the tools of network science should be used to support production flow analysis. Firstly, it was shown that the analysis of the paths in the network provides useful information about hidden, previously undefined connections. It was recognized that modularity analysis of the network is a promising tool for forming groups in PFA, and the performances of advanced (bipartite and multilayer) network modularity algorithms (like InfoMap) are comparable to the most advanced optimization algorithm tailored to the problem of cell formation.
A detailed benchmark problem was developed to make the research of multivariable algorithms of production flow analysis reproducible. With the help of the studied wire-harness process, the benefits of the modularity analysis of problem-specific sets of layers were demonstrated. The results confirm that the detected groups of activities are useful in terms of fine-tuning of modules (part families). Workload and capability-related network measures were developed. Along with analysis of the node degrees and their correlations, individual-, key-, and group-wise skills could be identified. The biadjacency matrices of the network lead to the calculation of workloads, and the investigation of how the production line is balanced. Besides the numerical analysis, visualizations were presented to demonstrate how multilayer networks provide insights into the critical factors of interconnected production systems, and the results of which confirm that multilayer networks can support the integration of production-relevant data and decision-making related to complex production systems.
Since the handling of the time-varying behaviour of process systems is becoming ever more critical in the field of cyber-physical systems, our future work will focus on the integration of historical process data to define networks of sequential procedures and temporal connections.