Abstract: Intelligence and servitization are important trends in the development of industrial modernization. Smart factory is the carrier to realize smart manufacturing, factory planning is the first step to build a smart factory, and the layout of production facilities is an important part of smart factory planning. The layout of production facilities has an important impact on the productivity, material handling efficiency, and production cycle of manufacturing systems. For the smart manufacturing cell system with resource coordination constraints, a quadratic assignment problem of facility layout to maximize the average throughput of the system was studied, and a queuing network based meta-heuristic optimization algorithm was proposed. Due to the stochastic property of the system under the customized production mode, an open queuing network model with blocking and generally distributed service time was established, and an approximate solution method was proposed to evaluate the system throughput. Then, a non-linear stochastic programming model for the facility layout problem was developed, and an improved variable neighborhood search algorithm based on queueing network was proposed to solve the solutions. In the variable neighborhood search algorithm, the initial solution generation method based on the processing route was improved, and the scale of the solution space of the neighborhood search was reduced by the isomorphism determination of directed graphs, to improve the efficiency of the algorithm. The comparison of experiments proved that our proposed algorithm has obvious advantages in both the quality of the optimal solution and the efficiency of the solution. Experiments show that the improved R-VNS* and D-VNS* algorithms have similar solution results due to the improved initial solution generation method, which avoids getting trapped in local optimum, and both of them outperform the improved R-VNS and D-VNS algorithms before the improvement. Due to the improvement of the initial solution generation method and the strategy of eliminating the homomorphic coding to reduce the solution space, both the improved R-VNS* and D-VNS* algorithms have improved the algorithm time complexity compared with the pre-improved algorithms, and the proposed R-VNS* algorithm based on the reduced-neighborhood search strategy operates more efficiently.

Key words: facility layout, quadratic assignment problem (QAP), queueing network, variable neighborhood search (VNS)