I suggest to have a look at the paper "An agent-baesd approach to self-organized production", see http://link.springer.com/chapter/10.1007/978-3-540-74089-6_7#page-1 or http://arxiv.org/pdf/1012.4645v1.pdf
I hope this is useful for you. Best regards, Dirk Helbing
Prof Flavio Tonelli : Yup, i'm thinking to model and simulate the interaction between people, equipment and lean tools ...the aims is for reduction or quantifying waste..."simulation-->agent based-->lean tools--->waste"
I just acquired Anylogic 7 and I agree with Flavio that it is a good approach for your scenario. You can model the people, equipment, and tools as agents.
I think ARENA can provide a perfect support to modelling and simulating the production process ... if you are going to evaluate an evolutionary improvement of this process accordingly to the level of best practices adoption (such as SMED and 5S and Kanban) by the people in their individual/aggregate dimension you definitely need some hybrid approach to blend discrete events with agents.
On the other side if you just assume the process has to be evaluated after some best practice is introduced (the assumption is about the perfect adoption of it by the organisation) you can simulate the process in the new conditions just using ARENA ... but in this way there is no "agents" usefulness since you work with scenarios.
I have been involved in manufacturing simulation for research in more sophisticated manufacturing execution than lean (e.g. incorporating a production radar service generating predictions how the system will behave while disturbances occur).
If you need a ready-to-use tool, anylogic is an option that allows you to do quite a lot (but perhaps too much so you can get lost in feature and option space especially when working in teams). The more conventional tools like arena, flexsim, ... have significant limitations, e.g. they do not support the production activity (an intelligent product) as a first class software entity. It are rules/policies/code/... attached to resources (machines) that take decisions. This mimics the manned production situation in which workers are co-located with machines while semi-finished products and parts only have a paper document accompanying them. In this organization, changes require retraining of the workers. In software, this implies software maintenance (and thus will never achieve flexible self-configurating production because software development/maintenance takes a lot of time and expertise).
To answer our requirements, we ended up using Erlang/OTP to program the MES and the factory/production emulation ourselves. It is superior to generic agent programming tools for this purpose. I am currently working on a software platform to be used by a community but it is going slowly (hard to get funding for research supporting activities that are not research themselves). Today, this option requires you to develop the software yourself (where Erlang coding will take 10-25 times less effort and time than c++). Cf. erlang.org.
I think any discrete event simulation tool may be appropriate if you can model the disturbances correctly. If you have one production flow to start with and measure the disturbances in time, quality etc., and then model how the disturbances change if you introduce 5S etc. (Implementation of SMED is easier, there you just change the input for setuptime). I think that if you measure those areas in your production line it might be possible to simulate accurate enough with simple DES tools to make a decision. The problem however is to visualise what you did to the management board for them to belive in the results. Then often more advanced tools are needed.
We have tried advanced vs simple DES-tools and the decision support were usually enough with the simple tools. Although the simulation engineer needs to be on top of which simplifications are made in the model.
In lean manufacturing a general rule is keep it simple and thus it is often a good idea t keep models as simple as possible.
If you have some experience on Arena, you should have a look at Simio (link 1). It uses many similar principles, plus it adds more intelligence on parts/items/products (moving entities). There is a guide especially for Arena users as well (link 2). We have used it for manufacturing and logistics scenarios with good results.
The software depends on your case. Lean simulations as we do them are currently E =Q/L =QFD/LCA based and typical simulation question of for instance Factor 4 = 4 times better customer service/4 times less loads (Muda/Muri) type. Typical Lean-line is tackt/standard time based so time based simiulation broblems are based on solving bottlenecks addressed within single kemba. As for instance Kai Zen involves two interacting bianry processes (Kai = change process, Zen is assessing what is achievable E). Simulation is thus on standardised work before the change vs. after and thus outcome is a development trend (delta). Now the problem becomes complex when the network is introduced as it involves incomplete, dynamic and systemic data (parakonsistent situation). As W3C cannot cope that kind of fuzzy data this requires open standardised work (web service) requests. Thus any current simulation tool with survival of fittest type algorithm will do. We are currently developing a more dynamic one based on parakonsitent logic and fuzzy data for self configuration of orders in lean situation but it requires more advanced internet protocol for network.
Agents could be useful if your lean manufacturing simulation includes assembly line workers and you want to include the variability of these workers.
Here is a book chapter I wrote (a long time ago) which might be of interest:
Siebers PO (2006) 'Worker Performance Modeling in Manufacturing Systems Simulation'. In: Rennard JP (Ed). Handbook of Research on Nature Inspired Computing for Economy and Management. Pennsylvania: Idea Group Publishing, pp. 661-678.