As it has been said, finding good solutions for the MMJIT problem with setups using an algorithmic approach may take too long and, on top of this, the solution can be vulnerable to product mix changes. Indeed, Kanban technique and GCM I methods are the most used approaches to manage JIT production thanks to their simplicity1. Some companies, where SMED techniques2 failed to reduce setup times, use a modified version of the kanban FIFO board, in order to prevent setups proliferation. Thus, a simple batching process is introduced: when more than one kanban is posted on the board, the workstation operator shall not start the job on the first row but, on the contrary, chooses the job which allows the workstation to skip the setup phase. As an example, given the original job sequence A-B-A-C-A-B for a workstation, if the operator is allowed to look two positions ahead, he would process A-A-B-C-A-B, saving one setup time. In such situations, where setup times cannot be reduced under a certain value, rather than giving up the idea of adopting the Lean Production approach, heuristics can be developed and tested in order to obtain a leveled production even if coping with long setup times or demand variability.
The most common method to analyze and validate heuristics is through simulation. Several authors agree that simulation is one of the best ways to analyze the dynamic and stochastic behavior of manufacturing system, predicting its operational performance3, 4, 5. Simulating, a user can dynamically reproduce how a system works and how the subsystems interact between each other; on top of this, a simulation tool can be used as a decision support system tool since it natively embeds the what-if logic6. Indeed, simulation can be used to test the solutions provided by Genetic Algorithms, Simulated Annealing, Ant Colony, etc. since these algorithms handle stochasticity and do not assume determinism. Simulation can be used for:
In spite of these potentialities, there seem to be few manufacturing simulation software really intended for industrial use, which go beyond a simple representation of the plant layout and modeling of the manufacturing flow. On top of some customized simulators – developed and built in a high-level programming language from some academic or research group in order to solve specific cases with drastic simplifying hypotheses – the major part of commercial software implements a graphical model-building approach, where experienced users can model almost any type of process using basic function blocks and evaluate the whole system behavior through some user-defined statistical functions14. The latters, being multi-purpose simulation software, require great efforts in translating real industrial processes logic into the modeling scheme, and it is thus difficult to “put down the simulation in the manufacturing process”15. Indeed, the lack of manufacturing archetypes to model building seems one of the most remarkable weakness for most simulator tools, since their presence could simplify the model development process for who speak the "language of business"16. Moreover, commercial simulators show several limitations if used to test custom heuristics, for example to level a JIT production or to solve a line-balancing problem: some authors report typical weaknesses in presenting the simulation output17 or limited functionalities in terms of statistical analysis18, on top of the lack of user-friendliness. For instance, most common commercial simulation software do not embed the most useful random distributions for manufacturing system analysis, such as the Weibull, Beta and Poisson distribution. When dealing with these cases, it is often easier to build custom software, despite it requires strong competences in operations research or statistics that have never represented the traditional background of industrial companies analysts19.
In order to widespread simulation software usage among the manufacturing industry, some authors underline the need of a standard architecture to model production and logistics processes20, 21, 22. Literature suggested to focus on a new reference framework for manufacturing simulation systems, that implement both a structure and a logic closer to real production systems and that may support industrial processes optimization23, 24.
Moreover, given hardware increased performances, computational workload of a simulation tool is not a problem anymore25 and it seems possible to develop simulators able to run in less than one minute even complex instances. The complexity of a manufacturing model is linked both to size and system stochasticity. A careful analysis of time series can provide useful information to be included in the simulator, in order to model stochastic variables linked to machine failures or scrap production. This allows a more truthful assessment of key performance indicators (KPI) for a range of solutions under test.
- 1146 reads
