What is the purpose of reinforcement learning in FlexSim?

[qhrrl4556]

[ FlexSim 22.2.2 ]

* Notice that this article was written using a translator

Hello, I am a developer and project manager working for a Korean manufacturing company.

Recently, while working on flexsim-related tasks, I became interested in using flexsim and reinforcement learning together.

In the flexsim example, I followed the example using gym, and when I was trying to figure out how to apply it as a real job, I got a question.

First, I wonder if my understanding is correct. My understanding of the content of tutorial using gym was as follows.

-The purpose is to create an artificial intelligence model(RL) that determines which item to send to the processor among the items (1~5) accumulated in the queue to increase the throughput the most.

I think that simulation, which I understand, is used to verify in advance if there is a wrong process before applying a real world problem.

However, I wondered what problems in reality would be solved by making the strategy of transferring from Queue to Processor in the simulation into a reinforcement learning model.

So.

1. I wonder what kind of value in real world we can develop into by developing the tutorial

2. What keywords are there to look for problems similar to optimization problems that Gym tutorial wants to solve

ex/ combination optimization, job shop scheduling... etc

3. Would you like to develop more examples of reinforcement learning with flexsim?

p.s. I'm very excited about FlexSim's reinforcement learning capabilities, and I'm looking forward to it. I hope that content related to reinforcement learning will continue to be added in the next version : ) thx for reading

[Jeanette_Fullmer]

Hello @Qhrrl4556,

Thank you for your great questions. Reinforcement Learning is used to create a brain that can train in a simulation and then be applied in the real world. For instance the tutorial model will develop a brain to pick the best items to send to the processor. However, this brain will not give you set rules to abide by like optimization. It is given inputs, makes a decision, and gives an output.

Here is a GitHub Repo that has more tutorials and if you haven't seen our manual yet here is the link to that as well.

[JordanLJohnson · Accepted solution]

1) Your understanding of the first tutorial model is correct. The brain learns how to choose items to improve throughput. In the tutorial example, the AI learns about a simple system, so it may not seem very useful. However, the same concept could be applied to a much more complicated system. There is very likely a real production system where having an AI choose which item to process next could significantly improve real throughput.

It turns out that correctly identifying a problem where AI would be helpful is a difficult problem. Usually, you want to find a person or policy in your organization who is making a decision frequently, and where that decision has a big impact on the system's performance. Then you can train an AI to make that decision, to aid the person in that decision, or to free them up for other tasks.

2) The keywords you have suggested seem like excellent starting points. Consider also including the term Discrete Event Simulation, so that your search will find AIs trained with a Discrete Event Simulation package.

3) Yes, we are interested in making more samples. It will likely take a while. As we gain experience using FlexSim with AI, we'll be able to create better and better examples.

[qhrrl4556]

thx for comment : )