In complex production environments, as in the automotive industry, machine schedules have to be determined under frequently changing customer demands, possible production failures, or unplanned delays. To determine even an almost optimal schedule in such an environment is a challenge. The reasons are the high complexity of the scheduling problem and uncertainty of several system parameters.
This paper presents a comprehensive methodology for production scheduling optimization within the scope of uncertainty of production parameters. These parameters are taken from a real application case of a supplier in the automotive industry. The research describes a two-stage scheduling problem with unrelated machines, machine qualifications, and skipping stages (a two-stage hybrid flow shop).
About the simulation model
To evaluate several optimization algorithms for production scheduling, the researchers built a detailed simulation model of the flow shop using AnyLogic software 8.5. By means of simulation, it was possible to include the interdependence of production parameters, to evaluate and potentially improve schedules before they were applied in the real production environment. The research team could also select the most suitable schedules depending on various demand levels.
The major source of information about the hybrid flow shop model parameters was the available data from the running system. In production systems, often a large amount of data is available and can be used to model uncertainty appropriately.
In this research, the team distinguished between internal parameters of the production system (processing or setup times, scrap rates, and availability of machines) and external parameters (mainly the varying demand).
In this paper, the researchers presented a simulation-based optimization solution developed with AnyLogic software. The aim was to solve production scheduling problems under uncertainty, which had been derived from a real industrial use case.
The production scheduling solution shows that the combination of data analysis to estimate the flow shop model parameters, heuristic optimization, and detailed stochastic simulation results in robust schedules that can be used in real systems.